首次提交

get_caption_mask.py

+import time
+import numpy as np
+
+from transformers import BertTokenizer
+from groundingdino.models.GroundingDINO.bertwarper import generate_masks_with_special_tokens_and_transfer_map
+
+# 输入你的文本提示词
+caption = 'car .'
+
+captions = [caption]
+# 编码文本
+# 使用模型的 tokenizer 对 caption 进行分词，并将其转换为张量格式
+t0 = time.time()
+tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+print(f"Loaded BERT tokenizer took {(time.time() - t0):.3f}s")
+t0 = time.time()
+tokenized = tokenizer(captions, padding="longest", return_tensors="pt").to('cpu')  # padding="longest" 确保在批处理中对齐较短的句子
+specical_tokens = tokenizer.convert_tokens_to_ids (["[CLS]", "[SEP]", ".", "?"])  # 将特殊字符（如 [CLS]、[SEP] 等）转换为它们在词汇表中的对应 ID
+print(f"Word embedding took {(time.time() - t0):.3f}s")
+
+# 生成注意力掩码和位置信息
+# 生成自注意力掩码，位置信息和类别到 token 的映射。这些掩码用于在 Transformer 中对注意力进行控制
+t0 = time.time()
+(
+    text_self_attention_masks,
+    position_ids,
+    cate_to_token_mask_list,
+) = generate_masks_with_special_tokens_and_transfer_map(
+    tokenized, specical_tokens, tokenizer)
+print(f"Generate attention masks took {(time.time() - t0):.3f}s")
+
+# 处理超长文本
+max_text_len = 256
+# 如果 caption 的长度超过模型的最大长度 max_text_len，则进行裁剪处理，包括裁剪输入 ID、注意力掩码和 token 类型 ID
+if text_self_attention_masks.shape[1] > max_text_len:
+    text_self_attention_masks = text_self_attention_masks[
+                                :, : max_text_len, : max_text_len]
+
+    position_ids = position_ids[:, : max_text_len]
+    tokenized["input_ids"] = tokenized["input_ids"][:, : max_text_len]
+    tokenized["attention_mask"] = tokenized["attention_mask"][:, : max_text_len]
+    tokenized["token_type_ids"] = tokenized["token_type_ids"][:, : max_text_len]
+
+
+attention_mask = np.asarray(tokenized["attention_mask"]).astype(bool)
+input_dict = {"input_ids": np.asarray(tokenized["input_ids"]), "attention_mask": attention_mask,
+         "position_ids": np.asarray(position_ids), "token_type_ids": np.asarray(tokenized["token_type_ids"]), "text_token_mask": np.asarray(text_self_attention_masks)}
+
+print(input_dict)
\ No newline at end of file

groundingdino/config/GroundingDINO_SwinB_cfg.py

+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_B_384_22k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

groundingdino/config/GroundingDINO_SwinT_OGC.py

+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_T_224_1k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

groundingdino/datasets/__init__.py

+

groundingdino/datasets/cocogrounding_eval.py

+# ------------------------------------------------------------------------
+# Grounding DINO. Midified by Shilong Liu.
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copyright (c) Aishwarya Kamath & Nicolas Carion. Licensed under the Apache License 2.0. All Rights Reserved
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+"""
+COCO evaluator that works in distributed mode.
+
+Mostly copy-paste from https://github.com/pytorch/vision/blob/edfd5a7/references/detection/coco_eval.py
+The difference is that there is less copy-pasting from pycocotools
+in the end of the file, as python3 can suppress prints with contextlib
+"""
+import contextlib
+import copy
+import os
+
+import numpy as np
+import pycocotools.mask as mask_util
+import torch
+from pycocotools.coco import COCO
+from pycocotools.cocoeval import COCOeval
+
+from groundingdino.util.misc import all_gather
+
+
+class CocoGroundingEvaluator(object):
+    def __init__(self, coco_gt, iou_types, useCats=True):
+        assert isinstance(iou_types, (list, tuple))
+        coco_gt = copy.deepcopy(coco_gt)
+        self.coco_gt = coco_gt
+
+        self.iou_types = iou_types
+        self.coco_eval = {}
+        for iou_type in iou_types:
+            self.coco_eval[iou_type] = COCOeval(coco_gt, iouType=iou_type)
+            self.coco_eval[iou_type].useCats = useCats
+
+        self.img_ids = []
+        self.eval_imgs = {k: [] for k in iou_types}
+        self.useCats = useCats
+
+    def update(self, predictions):
+        img_ids = list(np.unique(list(predictions.keys())))
+        self.img_ids.extend(img_ids)
+
+        for iou_type in self.iou_types:
+            results = self.prepare(predictions, iou_type)
+
+            # suppress pycocotools prints
+            with open(os.devnull, "w") as devnull:
+                with contextlib.redirect_stdout(devnull):
+                    coco_dt = COCO.loadRes(self.coco_gt, results) if results else COCO()
+
+            coco_eval = self.coco_eval[iou_type]
+
+            coco_eval.cocoDt = coco_dt
+            coco_eval.params.imgIds = list(img_ids)
+            coco_eval.params.useCats = self.useCats
+            img_ids, eval_imgs = evaluate(coco_eval)
+
+            self.eval_imgs[iou_type].append(eval_imgs)
+
+    def synchronize_between_processes(self):
+        for iou_type in self.iou_types:
+            self.eval_imgs[iou_type] = np.concatenate(self.eval_imgs[iou_type], 2)
+            create_common_coco_eval(self.coco_eval[iou_type], self.img_ids, self.eval_imgs[iou_type])
+
+    def accumulate(self):
+        for coco_eval in self.coco_eval.values():
+            coco_eval.accumulate()
+
+    def summarize(self):
+        for iou_type, coco_eval in self.coco_eval.items():
+            print("IoU metric: {}".format(iou_type))
+            coco_eval.summarize()
+
+    def prepare(self, predictions, iou_type):
+        if iou_type == "bbox":
+            return self.prepare_for_coco_detection(predictions)
+        elif iou_type == "segm":
+            return self.prepare_for_coco_segmentation(predictions)
+        elif iou_type == "keypoints":
+            return self.prepare_for_coco_keypoint(predictions)
+        else:
+            raise ValueError("Unknown iou type {}".format(iou_type))
+
+    def prepare_for_coco_detection(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "bbox": box,
+                        "score": scores[k],
+                    }
+                    for k, box in enumerate(boxes)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_segmentation(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            scores = prediction["scores"]
+            labels = prediction["labels"]
+            masks = prediction["masks"]
+
+            masks = masks > 0.5
+
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            rles = [
+                mask_util.encode(np.array(mask[0, :, :, np.newaxis], dtype=np.uint8, order="F"))[0] 
+                for mask in masks
+            ]
+            for rle in rles:
+                rle["counts"] = rle["counts"].decode("utf-8")
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "segmentation": rle,
+                        "score": scores[k],
+                    }
+                    for k, rle in enumerate(rles)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_keypoint(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+            keypoints = prediction["keypoints"]
+            keypoints = keypoints.flatten(start_dim=1).tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "keypoints": keypoint,
+                        "score": scores[k],
+                    }
+                    for k, keypoint in enumerate(keypoints)
+                ]
+            )
+        return coco_results
+
+
+def convert_to_xywh(boxes):
+    xmin, ymin, xmax, ymax = boxes.unbind(1)
+    return torch.stack((xmin, ymin, xmax - xmin, ymax - ymin), dim=1)
+
+
+def merge(img_ids, eval_imgs):
+    all_img_ids = all_gather(img_ids)
+    all_eval_imgs = all_gather(eval_imgs)
+
+    merged_img_ids = []
+    for p in all_img_ids:
+        merged_img_ids.extend(p)
+
+    merged_eval_imgs = []
+    for p in all_eval_imgs:
+        merged_eval_imgs.append(p)
+
+    merged_img_ids = np.array(merged_img_ids)
+    merged_eval_imgs = np.concatenate(merged_eval_imgs, 2)
+
+    # keep only unique (and in sorted order) images
+    merged_img_ids, idx = np.unique(merged_img_ids, return_index=True)
+    merged_eval_imgs = merged_eval_imgs[..., idx]
+
+    return merged_img_ids, merged_eval_imgs
+
+
+def create_common_coco_eval(coco_eval, img_ids, eval_imgs):
+    img_ids, eval_imgs = merge(img_ids, eval_imgs)
+    img_ids = list(img_ids)
+    eval_imgs = list(eval_imgs.flatten())
+
+    coco_eval.evalImgs = eval_imgs
+    coco_eval.params.imgIds = img_ids
+    coco_eval._paramsEval = copy.deepcopy(coco_eval.params)
+
+
+#################################################################
+# From pycocotools, just removed the prints and fixed
+# a Python3 bug about unicode not defined
+#################################################################
+
+
+def evaluate(self):
+    """
+    Run per image evaluation on given images and store results (a list of dict) in self.evalImgs
+    :return: None
+    """
+    # tic = time.time()
+    # print('Running per image evaluation...')
+    p = self.params
+    # add backward compatibility if useSegm is specified in params
+    if p.useSegm is not None:
+        p.iouType = "segm" if p.useSegm == 1 else "bbox"
+        print("useSegm (deprecated) is not None. Running {} evaluation".format(p.iouType))
+    # print('Evaluate annotation type *{}*'.format(p.iouType))
+    p.imgIds = list(np.unique(p.imgIds))
+    if p.useCats:
+        p.catIds = list(np.unique(p.catIds))
+    p.maxDets = sorted(p.maxDets)
+    self.params = p
+
+    self._prepare()
+    # loop through images, area range, max detection number
+    catIds = p.catIds if p.useCats else [-1]
+
+    if p.iouType == "segm" or p.iouType == "bbox":
+        computeIoU = self.computeIoU
+    elif p.iouType == "keypoints":
+        computeIoU = self.computeOks
+    self.ious = {
+        (imgId, catId): computeIoU(imgId, catId) 
+        for imgId in p.imgIds 
+        for catId in catIds}
+
+    evaluateImg = self.evaluateImg
+    maxDet = p.maxDets[-1]
+    evalImgs = [
+        evaluateImg(imgId, catId, areaRng, maxDet) 
+        for catId in catIds 
+        for areaRng in p.areaRng 
+        for imgId in p.imgIds
+    ]
+    # this is NOT in the pycocotools code, but could be done outside
+    evalImgs = np.asarray(evalImgs).reshape(len(catIds), len(p.areaRng), len(p.imgIds))
+    self._paramsEval = copy.deepcopy(self.params)
+    # toc = time.time()
+    # print('DONE (t={:0.2f}s).'.format(toc-tic))
+    return p.imgIds, evalImgs
+
+
+#################################################################
+# end of straight copy from pycocotools, just removing the prints
+#################################################################

groundingdino/datasets/cocogrounding_eval_zhongxing.py

+# ------------------------------------------------------------------------
+# Grounding DINO. Midified by Shilong Liu.
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copyright (c) Aishwarya Kamath & Nicolas Carion. Licensed under the Apache License 2.0. All Rights Reserved
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+"""
+COCO evaluator that works in distributed mode.
+
+Mostly copy-paste from https://github.com/pytorch/vision/blob/edfd5a7/references/detection/coco_eval.py
+The difference is that there is less copy-pasting from pycocotools
+in the end of the file, as python3 can suppress prints with contextlib
+"""
+import contextlib
+import copy
+import os
+
+import numpy as np
+import pycocotools.mask as mask_util
+import torch
+from pycocotools.coco import COCO
+from pycocotools.cocoeval import COCOeval
+
+from groundingdino.util.misc import all_gather
+
+
+class CocoGroundingEvaluator(object):
+    def __init__(self, coco_gt, iou_types, useCats=True):
+        assert isinstance(iou_types, (list, tuple))
+        coco_gt = copy.deepcopy(coco_gt)
+        self.coco_gt = coco_gt
+
+        self.iou_types = iou_types
+        self.coco_eval = {}
+        for iou_type in iou_types:
+            self.coco_eval[iou_type] = COCOeval(coco_gt, iouType=iou_type)
+            self.coco_eval[iou_type].useCats = useCats
+
+        self.img_ids = []
+        self.eval_imgs = {k: [] for k in iou_types}
+        self.useCats = useCats
+
+    def update(self, predictions):
+        img_ids = list(np.unique(list(predictions.keys())))
+        self.img_ids.extend(img_ids)
+        # import pdb;pdb.set_trace()
+        for iou_type in self.iou_types:
+            results = self.prepare(predictions, iou_type)
+
+            # suppress pycocotools prints
+            with open(os.devnull, "w") as devnull:
+                with contextlib.redirect_stdout(devnull):
+                    coco_dt = COCO.loadRes(self.coco_gt, results) if results else COCO()
+
+            coco_eval = self.coco_eval[iou_type]
+
+            coco_eval.cocoDt = coco_dt
+            coco_eval.params.imgIds = list(img_ids)
+            coco_eval.params.useCats = self.useCats
+            img_ids, eval_imgs = evaluate(coco_eval)
+
+            self.eval_imgs[iou_type].append(eval_imgs)
+
+    def synchronize_between_processes(self):
+        for iou_type in self.iou_types:
+            self.eval_imgs[iou_type] = np.concatenate(self.eval_imgs[iou_type], 2)
+            create_common_coco_eval(self.coco_eval[iou_type], self.img_ids, self.eval_imgs[iou_type])
+
+    def accumulate(self):
+        for coco_eval in self.coco_eval.values():
+            coco_eval.accumulate()
+
+    def summarize(self):
+        for iou_type, coco_eval in self.coco_eval.items():
+            print("IoU metric: {}".format(iou_type))
+            coco_eval.summarize()
+
+    def prepare(self, predictions, iou_type):
+        if iou_type == "bbox":
+            return self.prepare_for_coco_detection(predictions)
+        elif iou_type == "segm":
+            return self.prepare_for_coco_segmentation(predictions)
+        elif iou_type == "keypoints":
+            return self.prepare_for_coco_keypoint(predictions)
+        else:
+            raise ValueError("Unknown iou type {}".format(iou_type))
+
+    def prepare_for_coco_detection(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "bbox": box,
+                        "score": scores[k],
+                    }
+                    for k, box in enumerate(boxes)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_segmentation(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            scores = prediction["scores"]
+            labels = prediction["labels"]
+            masks = prediction["masks"]
+
+            masks = masks > 0.5
+
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            rles = [
+                mask_util.encode(np.array(mask[0, :, :, np.newaxis], dtype=np.uint8, order="F"))[0] 
+                for mask in masks
+            ]
+            for rle in rles:
+                rle["counts"] = rle["counts"].decode("utf-8")
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "segmentation": rle,
+                        "score": scores[k],
+                    }
+                    for k, rle in enumerate(rles)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_keypoint(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+            keypoints = prediction["keypoints"]
+            keypoints = keypoints.flatten(start_dim=1).tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "keypoints": keypoint,
+                        "score": scores[k],
+                    }
+                    for k, keypoint in enumerate(keypoints)
+                ]
+            )
+        return coco_results
+
+
+def convert_to_xywh(boxes):
+    xmin, ymin, xmax, ymax = boxes.unbind(1)
+    return torch.stack((xmin, ymin, xmax - xmin, ymax - ymin), dim=1)
+
+
+def merge(img_ids, eval_imgs):
+    all_img_ids = all_gather(img_ids)
+    all_eval_imgs = all_gather(eval_imgs)
+
+    merged_img_ids = []
+    for p in all_img_ids:
+        merged_img_ids.extend(p)
+
+    merged_eval_imgs = []
+    for p in all_eval_imgs:
+        merged_eval_imgs.append(p)
+
+    merged_img_ids = np.array(merged_img_ids)
+    merged_eval_imgs = np.concatenate(merged_eval_imgs, 2)
+
+    # keep only unique (and in sorted order) images
+    merged_img_ids, idx = np.unique(merged_img_ids, return_index=True)
+    merged_eval_imgs = merged_eval_imgs[..., idx]
+
+    return merged_img_ids, merged_eval_imgs
+
+
+def create_common_coco_eval(coco_eval, img_ids, eval_imgs):
+    img_ids, eval_imgs = merge(img_ids, eval_imgs)
+    img_ids = list(img_ids)
+    eval_imgs = list(eval_imgs.flatten())
+
+    coco_eval.evalImgs = eval_imgs
+    coco_eval.params.imgIds = img_ids
+    coco_eval._paramsEval = copy.deepcopy(coco_eval.params)
+
+
+#################################################################
+# From pycocotools, just removed the prints and fixed
+# a Python3 bug about unicode not defined
+#################################################################
+
+
+def evaluate(self):
+    """
+    Run per image evaluation on given images and store results (a list of dict) in self.evalImgs
+    :return: None
+    """
+    # tic = time.time()
+    # print('Running per image evaluation...')
+
+    # import pdb;pdb.set_trace()
+    p = self.params
+    # add backward compatibility if useSegm is specified in params
+    if p.useSegm is not None:
+        p.iouType = "segm" if p.useSegm == 1 else "bbox"
+        print("useSegm (deprecated) is not None. Running {} evaluation".format(p.iouType))
+    # print('Evaluate annotation type *{}*'.format(p.iouType))
+    p.imgIds = list(np.unique(p.imgIds))
+    if p.useCats:
+        p.catIds = list(np.unique(p.catIds))
+    p.maxDets = sorted(p.maxDets)
+    self.params = p
+
+    self._prepare()
+    # loop through images, area range, max detection number
+    catIds = p.catIds if p.useCats else [-1]
+
+    if p.iouType == "segm" or p.iouType == "bbox":
+        computeIoU = self.computeIoU
+    elif p.iouType == "keypoints":
+        computeIoU = self.computeOks
+    self.ious = {
+        (imgId, catId): computeIoU(imgId, catId) 
+        for imgId in p.imgIds 
+        for catId in catIds}
+
+    evaluateImg = self.evaluateImg
+    maxDet = p.maxDets[-1]
+    evalImgs = [
+        evaluateImg(imgId, catId, areaRng, maxDet) 
+        for catId in catIds 
+        for areaRng in p.areaRng 
+        for imgId in p.imgIds
+    ]
+    # this is NOT in the pycocotools code, but could be done outside
+    evalImgs = np.asarray(evalImgs).reshape(len(catIds), len(p.areaRng), len(p.imgIds))
+    self._paramsEval = copy.deepcopy(self.params)
+    # toc = time.time()
+    # print('DONE (t={:0.2f}s).'.format(toc-tic))
+    return p.imgIds, evalImgs
+
+
+#################################################################
+# end of straight copy from pycocotools, just removing the prints
+#################################################################

groundingdino/datasets/transforms.py

+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+"""
+Transforms and data augmentation for both image + bbox.
+"""
+import os
+import random
+
+import PIL
+import torch
+import torchvision.transforms as T
+import torchvision.transforms.functional as F
+
+from groundingdino.util.box_ops import box_xyxy_to_cxcywh
+from groundingdino.util.misc import interpolate
+
+
+def crop(image, target, region):
+    cropped_image = F.crop(image, *region)
+
+    target = target.copy()
+    i, j, h, w = region
+
+    # should we do something wrt the original size?
+    target["size"] = torch.tensor([h, w])
+
+    fields = ["labels", "area", "iscrowd", "positive_map"]
+
+    if "boxes" in target:
+        boxes = target["boxes"]
+        max_size = torch.as_tensor([w, h], dtype=torch.float32)
+        cropped_boxes = boxes - torch.as_tensor([j, i, j, i])
+        cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size)
+        cropped_boxes = cropped_boxes.clamp(min=0)
+        area = (cropped_boxes[:, 1, :] - cropped_boxes[:, 0, :]).prod(dim=1)
+        target["boxes"] = cropped_boxes.reshape(-1, 4)
+        target["area"] = area
+        fields.append("boxes")
+
+    if "masks" in target:
+        # FIXME should we update the area here if there are no boxes?
+        target["masks"] = target["masks"][:, i : i + h, j : j + w]
+        fields.append("masks")
+
+    # remove elements for which the boxes or masks that have zero area
+    if "boxes" in target or "masks" in target:
+        # favor boxes selection when defining which elements to keep
+        # this is compatible with previous implementation
+        if "boxes" in target:
+            cropped_boxes = target["boxes"].reshape(-1, 2, 2)
+            keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1)
+        else:
+            keep = target["masks"].flatten(1).any(1)
+
+        for field in fields:
+            if field in target:
+                target[field] = target[field][keep]
+
+    if os.environ.get("IPDB_SHILONG_DEBUG", None) == "INFO":
+        # for debug and visualization only.
+        if "strings_positive" in target:
+            target["strings_positive"] = [
+                _i for _i, _j in zip(target["strings_positive"], keep) if _j
+            ]
+
+    return cropped_image, target
+
+
+def hflip(image, target):
+    flipped_image = F.hflip(image)
+
+    w, h = image.size
+
+    target = target.copy()
+    if "boxes" in target:
+        boxes = target["boxes"]
+        boxes = boxes[:, [2, 1, 0, 3]] * torch.as_tensor([-1, 1, -1, 1]) + torch.as_tensor(
+            [w, 0, w, 0]
+        )
+        target["boxes"] = boxes
+
+    if "masks" in target:
+        target["masks"] = target["masks"].flip(-1)
+
+    return flipped_image, target
+
+
+def resize(image, target, size, max_size=None):
+    # size can be min_size (scalar) or (w, h) tuple
+
+    def get_size_with_aspect_ratio(image_size, size, max_size=None):
+        w, h = image_size
+        if max_size is not None:
+            min_original_size = float(min((w, h)))
+            max_original_size = float(max((w, h)))
+            if max_original_size / min_original_size * size > max_size:
+                size = int(round(max_size * min_original_size / max_original_size))
+
+        if (w <= h and w == size) or (h <= w and h == size):
+            return (h, w)
+
+        if w < h:
+            ow = size
+            oh = int(size * h / w)
+        else:
+            oh = size
+            ow = int(size * w / h)
+
+        return (oh, ow)
+
+    def get_size(image_size, size, max_size=None):
+        if isinstance(size, (list, tuple)):
+            return size[::-1]
+        else:
+            return get_size_with_aspect_ratio(image_size, size, max_size)
+
+    size = get_size(image.size, size, max_size)
+    rescaled_image = F.resize(image, size)
+
+    if target is None:
+        return rescaled_image, None
+
+    ratios = tuple(float(s) / float(s_orig) for s, s_orig in zip(rescaled_image.size, image.size))
+    ratio_width, ratio_height = ratios
+
+    target = target.copy()
+    if "boxes" in target:
+        boxes = target["boxes"]
+        scaled_boxes = boxes * torch.as_tensor(
+            [ratio_width, ratio_height, ratio_width, ratio_height]
+        )
+        target["boxes"] = scaled_boxes
+
+    if "area" in target:
+        area = target["area"]
+        scaled_area = area * (ratio_width * ratio_height)
+        target["area"] = scaled_area
+
+    h, w = size
+    target["size"] = torch.tensor([h, w])
+
+    if "masks" in target:
+        target["masks"] = (
+            interpolate(target["masks"][:, None].float(), size, mode="nearest")[:, 0] > 0.5
+        )
+
+    return rescaled_image, target
+
+
+def pad(image, target, padding):
+    # assumes that we only pad on the bottom right corners
+    padded_image = F.pad(image, (0, 0, padding[0], padding[1]))
+    if target is None:
+        return padded_image, None
+    target = target.copy()
+    # should we do something wrt the original size?
+    target["size"] = torch.tensor(padded_image.size[::-1])
+    if "masks" in target:
+        target["masks"] = torch.nn.functional.pad(target["masks"], (0, padding[0], 0, padding[1]))
+    return padded_image, target
+
+
+class ResizeDebug(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        return resize(img, target, self.size)
+
+
+class RandomCrop(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        region = T.RandomCrop.get_params(img, self.size)
+        return crop(img, target, region)
+
+
+class RandomSizeCrop(object):
+    def __init__(self, min_size: int, max_size: int, respect_boxes: bool = False):
+        # respect_boxes:    True to keep all boxes
+        #                   False to tolerence box filter
+        self.min_size = min_size
+        self.max_size = max_size
+        self.respect_boxes = respect_boxes
+
+    def __call__(self, img: PIL.Image.Image, target: dict):
+        init_boxes = len(target["boxes"])
+        max_patience = 10
+        for i in range(max_patience):
+            w = random.randint(self.min_size, min(img.width, self.max_size))
+            h = random.randint(self.min_size, min(img.height, self.max_size))
+            region = T.RandomCrop.get_params(img, [h, w])
+            result_img, result_target = crop(img, target, region)
+            if (
+                not self.respect_boxes
+                or len(result_target["boxes"]) == init_boxes
+                or i == max_patience - 1
+            ):
+                return result_img, result_target
+        return result_img, result_target
+
+
+class CenterCrop(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        image_width, image_height = img.size
+        crop_height, crop_width = self.size
+        crop_top = int(round((image_height - crop_height) / 2.0))
+        crop_left = int(round((image_width - crop_width) / 2.0))
+        return crop(img, target, (crop_top, crop_left, crop_height, crop_width))
+
+
+class RandomHorizontalFlip(object):
+    def __init__(self, p=0.5):
+        self.p = p
+
+    def __call__(self, img, target):
+        if random.random() < self.p:
+            return hflip(img, target)
+        return img, target
+
+
+class RandomResize(object):
+    def __init__(self, sizes, max_size=None):
+        assert isinstance(sizes, (list, tuple))
+        self.sizes = sizes
+        self.max_size = max_size
+
+    def __call__(self, img, target=None):
+        size = random.choice(self.sizes)
+        return resize(img, target, size, self.max_size)
+
+
+class RandomPad(object):
+    def __init__(self, max_pad):
+        self.max_pad = max_pad
+
+    def __call__(self, img, target):
+        pad_x = random.randint(0, self.max_pad)
+        pad_y = random.randint(0, self.max_pad)
+        return pad(img, target, (pad_x, pad_y))
+
+
+class RandomSelect(object):
+    """
+    Randomly selects between transforms1 and transforms2,
+    with probability p for transforms1 and (1 - p) for transforms2
+    """
+
+    def __init__(self, transforms1, transforms2, p=0.5):
+        self.transforms1 = transforms1
+        self.transforms2 = transforms2
+        self.p = p
+
+    def __call__(self, img, target):
+        if random.random() < self.p:
+            return self.transforms1(img, target)
+        return self.transforms2(img, target)
+
+
+class ToTensor(object):
+    def __call__(self, img, target):
+        return F.to_tensor(img), target
+
+
+class RandomErasing(object):
+    def __init__(self, *args, **kwargs):
+        self.eraser = T.RandomErasing(*args, **kwargs)
+
+    def __call__(self, img, target):
+        return self.eraser(img), target
+
+
+class Normalize(object):
+    def __init__(self, mean, std):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, image, target=None):
+        image = F.normalize(image, mean=self.mean, std=self.std)
+        if target is None:
+            return image, None
+        target = target.copy()
+        h, w = image.shape[-2:]
+        if "boxes" in target:
+            boxes = target["boxes"]
+            boxes = box_xyxy_to_cxcywh(boxes)
+            boxes = boxes / torch.tensor([w, h, w, h], dtype=torch.float32)
+            target["boxes"] = boxes
+        return image, target
+
+
+class Compose(object):
+    def __init__(self, transforms):
+        self.transforms = transforms
+
+    def __call__(self, image, target):
+        for t in self.transforms:
+            image, target = t(image, target)
+        return image, target
+
+    def __repr__(self):
+        format_string = self.__class__.__name__ + "("
+        for t in self.transforms:
+            format_string += "\n"
+            format_string += "    {0}".format(t)
+        format_string += "\n)"
+        return format_string

groundingdino/models/GroundingDINO/__init__.py

+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+from .groundingdino import build_groundingdino

groundingdino/models/GroundingDINO/backbone/__init__.py

+from .backbone import build_backbone

groundingdino/models/GroundingDINO/backbone/backbone.py

+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+"""
+Backbone modules.
+"""
+
+from typing import Dict, List
+
+import torch
+import torch.nn.functional as F
+import torchvision
+from torch import nn
+from torchvision.models._utils import IntermediateLayerGetter
+
+from groundingdino.util.misc import NestedTensor, clean_state_dict, is_main_process
+
+from .position_encoding import build_position_encoding
+from .swin_transformer import build_swin_transformer
+
+
+class FrozenBatchNorm2d(torch.nn.Module):
+    """
+    BatchNorm2d where the batch statistics and the affine parameters are fixed.
+
+    Copy-paste from torchvision.misc.ops with added eps before rqsrt,
+    without which any other models than torchvision.models.resnet[18,34,50,101]
+    produce nans.
+    """
+
+    def __init__(self, n):
+        super(FrozenBatchNorm2d, self).__init__()
+        self.register_buffer("weight", torch.ones(n))
+        self.register_buffer("bias", torch.zeros(n))
+        self.register_buffer("running_mean", torch.zeros(n))
+        self.register_buffer("running_var", torch.ones(n))
+
+    def _load_from_state_dict(
+        self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
+    ):
+        num_batches_tracked_key = prefix + "num_batches_tracked"
+        if num_batches_tracked_key in state_dict:
+            del state_dict[num_batches_tracked_key]
+
+        super(FrozenBatchNorm2d, self)._load_from_state_dict(
+            state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
+        )
+
+    def forward(self, x):
+        # move reshapes to the beginning
+        # to make it fuser-friendly
+        w = self.weight.reshape(1, -1, 1, 1)
+        b = self.bias.reshape(1, -1, 1, 1)
+        rv = self.running_var.reshape(1, -1, 1, 1)
+        rm = self.running_mean.reshape(1, -1, 1, 1)
+        eps = 1e-5
+        scale = w * (rv + eps).rsqrt()
+        bias = b - rm * scale
+        return x * scale + bias
+
+
+class BackboneBase(nn.Module):
+    def __init__(
+        self,
+        backbone: nn.Module,
+        train_backbone: bool,
+        num_channels: int,
+        return_interm_indices: list,
+    ):
+        super().__init__()
+        for name, parameter in backbone.named_parameters():
+            if (
+                not train_backbone
+                or "layer2" not in name
+                and "layer3" not in name
+                and "layer4" not in name
+            ):
+                parameter.requires_grad_(False)
+
+        return_layers = {}
+        for idx, layer_index in enumerate(return_interm_indices):
+            return_layers.update(
+                {"layer{}".format(5 - len(return_interm_indices) + idx): "{}".format(layer_index)}
+            )
+
+        # if len:
+        #     if use_stage1_feature:
+        #         return_layers = {"layer1": "0", "layer2": "1", "layer3": "2", "layer4": "3"}
+        #     else:
+        #         return_layers = {"layer2": "0", "layer3": "1", "layer4": "2"}
+        # else:
+        #     return_layers = {'layer4': "0"}
+        self.body = IntermediateLayerGetter(backbone, return_layers=return_layers)
+        self.num_channels = num_channels
+
+    def forward(self, tensor_list: NestedTensor):
+        xs = self.body(tensor_list.tensors)
+        out: Dict[str, NestedTensor] = {}
+        for name, x in xs.items():
+            m = tensor_list.mask
+            assert m is not None
+            mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0]
+            out[name] = NestedTensor(x, mask)
+        # import ipdb; ipdb.set_trace()
+        return out
+
+
+class Backbone(BackboneBase):
+    """ResNet backbone with frozen BatchNorm."""
+
+    def __init__(
+        self,
+        name: str,
+        train_backbone: bool,
+        dilation: bool,
+        return_interm_indices: list,
+        batch_norm=FrozenBatchNorm2d,
+    ):
+        if name in ["resnet18", "resnet34", "resnet50", "resnet101"]:
+            backbone = getattr(torchvision.models, name)(
+                replace_stride_with_dilation=[False, False, dilation],
+                pretrained=is_main_process(),
+                norm_layer=batch_norm,
+            )
+        else:
+            raise NotImplementedError("Why you can get here with name {}".format(name))
+        # num_channels = 512 if name in ('resnet18', 'resnet34') else 2048
+        assert name not in ("resnet18", "resnet34"), "Only resnet50 and resnet101 are available."
+        assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
+        num_channels_all = [256, 512, 1024, 2048]
+        num_channels = num_channels_all[4 - len(return_interm_indices) :]
+        super().__init__(backbone, train_backbone, num_channels, return_interm_indices)
+
+
+class Joiner(nn.Sequential):
+    def __init__(self, backbone, position_embedding):
+        super().__init__(backbone, position_embedding)
+
+    def forward(self, tensor_list: NestedTensor):
+        xs = self[0](tensor_list)
+        out: List[NestedTensor] = []
+        pos = []
+        for name, x in xs.items():
+            out.append(x)
+            # position encoding
+            pos.append(self[1](x).to(x.tensors.dtype))
+
+        return out, pos
+
+
+def build_backbone(args):
+    """
+    Useful args:
+        - backbone: backbone name
+        - lr_backbone:
+        - dilation
+        - return_interm_indices: available: [0,1,2,3], [1,2,3], [3]
+        - backbone_freeze_keywords:
+        - use_checkpoint: for swin only for now
+
+    """
+    position_embedding = build_position_encoding(args)
+    train_backbone = True
+    if not train_backbone:
+        raise ValueError("Please set lr_backbone > 0")
+    return_interm_indices = args.return_interm_indices
+    assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
+    args.backbone_freeze_keywords
+    use_checkpoint = getattr(args, "use_checkpoint", False)
+
+    if args.backbone in ["resnet50", "resnet101"]:
+        backbone = Backbone(
+            args.backbone,
+            train_backbone,
+            args.dilation,
+            return_interm_indices,
+            batch_norm=FrozenBatchNorm2d,
+        )
+        bb_num_channels = backbone.num_channels
+    elif args.backbone in [
+        "swin_T_224_1k",
+        "swin_B_224_22k",
+        "swin_B_384_22k",
+        "swin_L_224_22k",
+        "swin_L_384_22k",
+    ]:
+        pretrain_img_size = int(args.backbone.split("_")[-2])
+        backbone = build_swin_transformer(
+            args.backbone,
+            pretrain_img_size=pretrain_img_size,
+            out_indices=tuple(return_interm_indices),
+            dilation=False,
+            use_checkpoint=use_checkpoint,
+        )
+
+        bb_num_channels = backbone.num_features[4 - len(return_interm_indices) :]
+    else:
+        raise NotImplementedError("Unknown backbone {}".format(args.backbone))
+
+    assert len(bb_num_channels) == len(
+        return_interm_indices
+    ), f"len(bb_num_channels) {len(bb_num_channels)} != len(return_interm_indices) {len(return_interm_indices)}"
+
+    model = Joiner(backbone, position_embedding)
+    model.num_channels = bb_num_channels
+    assert isinstance(
+        bb_num_channels, List
+    ), "bb_num_channels is expected to be a List but {}".format(type(bb_num_channels))
+    # import ipdb; ipdb.set_trace()
+    return model

groundingdino/models/GroundingDINO/backbone/position_encoding.py

+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# DINO
+# Copyright (c) 2022 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+"""
+Various positional encodings for the transformer.
+"""
+import math
+
+import torch
+from torch import nn
+
+from groundingdino.util.misc import NestedTensor
+
+
+class PositionEmbeddingSine(nn.Module):
+    """
+    This is a more standard version of the position embedding, very similar to the one
+    used by the Attention is all you need paper, generalized to work on images.
+    """
+
+    def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None):
+        super().__init__()
+        self.num_pos_feats = num_pos_feats
+        self.temperature = temperature
+        self.normalize = normalize
+        if scale is not None and normalize is False:
+            raise ValueError("normalize should be True if scale is passed")
+        if scale is None:
+            scale = 2 * math.pi
+        self.scale = scale
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        mask = tensor_list.mask
+        assert mask is not None
+        not_mask = ~mask
+        y_embed = not_mask.cumsum(1, dtype=torch.float32)
+        x_embed = not_mask.cumsum(2, dtype=torch.float32)
+        if self.normalize:
+            eps = 1e-6
+            # if os.environ.get("SHILONG_AMP", None) == '1':
+            #     eps = 1e-4
+            # else:
+            #     eps = 1e-6
+            y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
+            x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
+
+        dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_t = self.temperature ** (2 * (dim_t // 2) / self.num_pos_feats)
+
+        pos_x = x_embed[:, :, :, None] / dim_t
+        pos_y = y_embed[:, :, :, None] / dim_t
+        pos_x = torch.stack(
+            (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos_y = torch.stack(
+            (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2)
+        return pos
+
+
+class PositionEmbeddingSineHW(nn.Module):
+    """
+    This is a more standard version of the position embedding, very similar to the one
+    used by the Attention is all you need paper, generalized to work on images.
+    """
+
+    def __init__(
+        self, num_pos_feats=64, temperatureH=10000, temperatureW=10000, normalize=False, scale=None
+    ):
+        super().__init__()
+        self.num_pos_feats = num_pos_feats
+        self.temperatureH = temperatureH
+        self.temperatureW = temperatureW
+        self.normalize = normalize
+        if scale is not None and normalize is False:
+            raise ValueError("normalize should be True if scale is passed")
+        if scale is None:
+            scale = 2 * math.pi
+        self.scale = scale
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        mask = tensor_list.mask
+        assert mask is not None
+        not_mask = ~mask
+        y_embed = not_mask.cumsum(1, dtype=torch.float32)
+        x_embed = not_mask.cumsum(2, dtype=torch.float32)
+
+        # import ipdb; ipdb.set_trace()
+
+        if self.normalize:
+            eps = 1e-6
+            y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
+            x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
+
+        dim_tx = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_tx = self.temperatureW ** (2 * (torch.div(dim_tx, 2, rounding_mode='floor')) / self.num_pos_feats)
+        pos_x = x_embed[:, :, :, None] / dim_tx
+
+        dim_ty = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_ty = self.temperatureH ** (2 * (torch.div(dim_ty, 2, rounding_mode='floor')) / self.num_pos_feats)
+        pos_y = y_embed[:, :, :, None] / dim_ty
+
+        pos_x = torch.stack(
+            (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos_y = torch.stack(
+            (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2)
+
+        # import ipdb; ipdb.set_trace()
+
+        return pos
+
+
+class PositionEmbeddingLearned(nn.Module):
+    """
+    Absolute pos embedding, learned.
+    """
+
+    def __init__(self, num_pos_feats=256):
+        super().__init__()
+        self.row_embed = nn.Embedding(50, num_pos_feats)
+        self.col_embed = nn.Embedding(50, num_pos_feats)
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        nn.init.uniform_(self.row_embed.weight)
+        nn.init.uniform_(self.col_embed.weight)
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        h, w = x.shape[-2:]
+        i = torch.arange(w, device=x.device)
+        j = torch.arange(h, device=x.device)
+        x_emb = self.col_embed(i)
+        y_emb = self.row_embed(j)
+        pos = (
+            torch.cat(
+                [
+                    x_emb.unsqueeze(0).repeat(h, 1, 1),
+                    y_emb.unsqueeze(1).repeat(1, w, 1),
+                ],
+                dim=-1,
+            )
+            .permute(2, 0, 1)
+            .unsqueeze(0)
+            .repeat(x.shape[0], 1, 1, 1)
+        )
+        return pos
+
+
+def build_position_encoding(args):
+    N_steps = args.hidden_dim // 2
+    if args.position_embedding in ("v2", "sine"):
+        # TODO find a better way of exposing other arguments
+        position_embedding = PositionEmbeddingSineHW(
+            N_steps,
+            temperatureH=args.pe_temperatureH,
+            temperatureW=args.pe_temperatureW,
+            normalize=True,
+        )
+    elif args.position_embedding in ("v3", "learned"):
+        position_embedding = PositionEmbeddingLearned(N_steps)
+    else:
+        raise ValueError(f"not supported {args.position_embedding}")
+
+    return position_embedding

groundingdino/models/GroundingDINO/backbone/swin_transformer.py

+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# DINO
+# Copyright (c) 2022 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# --------------------------------------------------------
+# modified from https://github.com/SwinTransformer/Swin-Transformer-Object-Detection/blob/master/mmdet/models/backbones/swin_transformer.py
+# --------------------------------------------------------
+
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint as checkpoint
+from timm.models.layers import DropPath, to_2tuple, trunc_normal_
+
+from groundingdino.util.misc import NestedTensor
+
+
+class Mlp(nn.Module):
+    """Multilayer perceptron."""
+
+    def __init__(
+        self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.0
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+def window_partition(x, window_size):
+    """
+    Args:
+        x: (B, H, W, C)
+        window_size (int): window size
+    Returns:
+        windows: (num_windows*B, window_size, window_size, C)
+    """
+    B, H, W, C = x.shape
+    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows
+
+
+def window_reverse(windows, window_size, H, W):
+    """
+    Args:
+        windows: (num_windows*B, window_size, window_size, C)
+        window_size (int): Window size
+        H (int): Height of image
+        W (int): Width of image
+    Returns:
+        x: (B, H, W, C)
+    """
+    B = int(windows.shape[0] / (H * W / window_size / window_size))
+    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
+    return x
+
+
+class WindowAttention(nn.Module):
+    """Window based multi-head self attention (W-MSA) module with relative position bias.
+    It supports both of shifted and non-shifted window.
+    Args:
+        dim (int): Number of input channels.
+        window_size (tuple[int]): The height and width of the window.
+        num_heads (int): Number of attention heads.
+        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
+        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
+        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
+    """
+
+    def __init__(
+        self,
+        dim,
+        window_size,
+        num_heads,
+        qkv_bias=True,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+
+        super().__init__()
+        self.dim = dim
+        self.window_size = window_size  # Wh, Ww
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        # define a parameter table of relative position bias
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
+        )  # 2*Wh-1 * 2*Ww-1, nH
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = torch.arange(self.window_size[0])
+        coords_w = torch.arange(self.window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+        trunc_normal_(self.relative_position_bias_table, std=0.02)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, mask=None):
+        """Forward function.
+        Args:
+            x: input features with shape of (num_windows*B, N, C)
+            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
+        """
+        B_, N, C = x.shape
+        qkv = (
+            self.qkv(x)
+            .reshape(B_, N, 3, self.num_heads, C // self.num_heads)
+            .permute(2, 0, 3, 1, 4)
+        )
+        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
+
+        q = q * self.scale
+        attn = q @ k.transpose(-2, -1)
+
+        relative_position_bias = self.relative_position_bias_table[
+            self.relative_position_index.view(-1)
+        ].view(
+            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1
+        )  # Wh*Ww,Wh*Ww,nH
+        relative_position_bias = relative_position_bias.permute(
+            2, 0, 1
+        ).contiguous()  # nH, Wh*Ww, Wh*Ww
+        attn = attn + relative_position_bias.unsqueeze(0)
+
+        if mask is not None:
+            nW = mask.shape[0]
+            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
+            attn = attn.view(-1, self.num_heads, N, N)
+            attn = self.softmax(attn)
+        else:
+            attn = self.softmax(attn)
+
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class SwinTransformerBlock(nn.Module):
+    """Swin Transformer Block.
+    Args:
+        dim (int): Number of input channels.
+        num_heads (int): Number of attention heads.
+        window_size (int): Window size.
+        shift_size (int): Shift size for SW-MSA.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float, optional): Stochastic depth rate. Default: 0.0
+        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        window_size=7,
+        shift_size=0,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+    ):
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
+
+        self.norm1 = norm_layer(dim)
+        self.attn = WindowAttention(
+            dim,
+            window_size=to_2tuple(self.window_size),
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(
+            in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop
+        )
+
+        self.H = None
+        self.W = None
+
+    def forward(self, x, mask_matrix):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+            mask_matrix: Attention mask for cyclic shift.
+        """
+        B, L, C = x.shape
+        H, W = self.H, self.W
+        assert L == H * W, "input feature has wrong size"
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(B, H, W, C)
+
+        # pad feature maps to multiples of window size
+        pad_l = pad_t = 0
+        pad_r = (self.window_size - W % self.window_size) % self.window_size
+        pad_b = (self.window_size - H % self.window_size) % self.window_size
+        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
+        _, Hp, Wp, _ = x.shape
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+            attn_mask = mask_matrix
+        else:
+            shifted_x = x
+            attn_mask = None
+
+        # partition windows
+        x_windows = window_partition(
+            shifted_x, self.window_size
+        )  # nW*B, window_size, window_size, C
+        x_windows = x_windows.view(
+            -1, self.window_size * self.window_size, C
+        )  # nW*B, window_size*window_size, C
+
+        # W-MSA/SW-MSA
+        attn_windows = self.attn(x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C
+
+        # merge windows
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
+        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C
+
+        # reverse cyclic shift
+        if self.shift_size > 0:
+            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
+        else:
+            x = shifted_x
+
+        if pad_r > 0 or pad_b > 0:
+            x = x[:, :H, :W, :].contiguous()
+
+        x = x.view(B, H * W, C)
+
+        # FFN
+        x = shortcut + self.drop_path(x)
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+
+        return x
+
+
+class PatchMerging(nn.Module):
+    """Patch Merging Layer
+    Args:
+        dim (int): Number of input channels.
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, dim, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
+        self.norm = norm_layer(4 * dim)
+
+    def forward(self, x, H, W):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+        B, L, C = x.shape
+        assert L == H * W, "input feature has wrong size"
+
+        x = x.view(B, H, W, C)
+
+        # padding
+        pad_input = (H % 2 == 1) or (W % 2 == 1)
+        if pad_input:
+            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
+
+        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
+        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
+        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
+        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
+        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
+        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
+
+        x = self.norm(x)
+        x = self.reduction(x)
+
+        return x
+
+
+class BasicLayer(nn.Module):
+    """A basic Swin Transformer layer for one stage.
+    Args:
+        dim (int): Number of feature channels
+        depth (int): Depths of this stage.
+        num_heads (int): Number of attention head.
+        window_size (int): Local window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
+        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
+        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+    """
+
+    def __init__(
+        self,
+        dim,
+        depth,
+        num_heads,
+        window_size=7,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        norm_layer=nn.LayerNorm,
+        downsample=None,
+        use_checkpoint=False,
+    ):
+        super().__init__()
+        self.window_size = window_size
+        self.shift_size = window_size // 2
+        self.depth = depth
+        self.use_checkpoint = use_checkpoint
+
+        # build blocks
+        self.blocks = nn.ModuleList(
+            [
+                SwinTransformerBlock(
+                    dim=dim,
+                    num_heads=num_heads,
+                    window_size=window_size,
+                    shift_size=0 if (i % 2 == 0) else window_size // 2,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop,
+                    attn_drop=attn_drop,
+                    drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
+                    norm_layer=norm_layer,
+                )
+                for i in range(depth)
+            ]
+        )
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
+        else:
+            self.downsample = None
+
+    def forward(self, x, H, W):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+
+        # calculate attention mask for SW-MSA
+        Hp = int(np.ceil(H / self.window_size)) * self.window_size
+        Wp = int(np.ceil(W / self.window_size)) * self.window_size
+        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
+        h_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        w_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        cnt = 0
+        for h in h_slices:
+            for w in w_slices:
+                img_mask[:, h, w, :] = cnt
+                cnt += 1
+
+        mask_windows = window_partition(
+            img_mask, self.window_size
+        )  # nW, window_size, window_size, 1
+        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
+        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(
+            attn_mask == 0, float(0.0)
+        )
+
+        for blk in self.blocks:
+            blk.H, blk.W = H, W
+            if self.use_checkpoint:
+                x = checkpoint.checkpoint(blk, x, attn_mask)
+            else:
+                x = blk(x, attn_mask)
+        if self.downsample is not None:
+            x_down = self.downsample(x, H, W)
+            Wh, Ww = (H + 1) // 2, (W + 1) // 2
+            return x, H, W, x_down, Wh, Ww
+        else:
+            return x, H, W, x, H, W
+
+
+class PatchEmbed(nn.Module):
+    """Image to Patch Embedding
+    Args:
+        patch_size (int): Patch token size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        norm_layer (nn.Module, optional): Normalization layer. Default: None
+    """
+
+    def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        patch_size = to_2tuple(patch_size)
+        self.patch_size = patch_size
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        """Forward function."""
+        # padding
+        _, _, H, W = x.size()
+        if W % self.patch_size[1] != 0:
+            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
+        if H % self.patch_size[0] != 0:
+            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))
+
+        x = self.proj(x)  # B C Wh Ww
+        if self.norm is not None:
+            Wh, Ww = x.size(2), x.size(3)
+            x = x.flatten(2).transpose(1, 2)
+            x = self.norm(x)
+            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
+
+        return x
+
+
+class SwinTransformer(nn.Module):
+    """Swin Transformer backbone.
+        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -
+          https://arxiv.org/pdf/2103.14030
+    Args:
+        pretrain_img_size (int): Input image size for training the pretrained model,
+            used in absolute postion embedding. Default 224.
+        patch_size (int | tuple(int)): Patch size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        depths (tuple[int]): Depths of each Swin Transformer stage.
+        num_heads (tuple[int]): Number of attention head of each stage.
+        window_size (int): Window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float): Override default qk scale of head_dim ** -0.5 if set.
+        drop_rate (float): Dropout rate.
+        attn_drop_rate (float): Attention dropout rate. Default: 0.
+        drop_path_rate (float): Stochastic depth rate. Default: 0.2.
+        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
+        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False.
+        patch_norm (bool): If True, add normalization after patch embedding. Default: True.
+        out_indices (Sequence[int]): Output from which stages.
+        frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
+            -1 means not freezing any parameters.
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+        dilation (bool): if True, the output size if 16x downsample, ow 32x downsample.
+    """
+
+    def __init__(
+        self,
+        pretrain_img_size=224,
+        patch_size=4,
+        in_chans=3,
+        embed_dim=96,
+        depths=[2, 2, 6, 2],
+        num_heads=[3, 6, 12, 24],
+        window_size=7,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.2,
+        norm_layer=nn.LayerNorm,
+        ape=False,
+        patch_norm=True,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=-1,
+        dilation=False,
+        use_checkpoint=False,
+    ):
+        super().__init__()
+
+        self.pretrain_img_size = pretrain_img_size
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.ape = ape
+        self.patch_norm = patch_norm
+        self.out_indices = out_indices
+        self.frozen_stages = frozen_stages
+        self.dilation = dilation
+
+        # if use_checkpoint:
+        #     print("use_checkpoint!!!!!!!!!!!!!!!!!!!!!!!!")
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed(
+            patch_size=patch_size,
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None,
+        )
+
+        # absolute position embedding
+        if self.ape:
+            pretrain_img_size = to_2tuple(pretrain_img_size)
+            patch_size = to_2tuple(patch_size)
+            patches_resolution = [
+                pretrain_img_size[0] // patch_size[0],
+                pretrain_img_size[1] // patch_size[1],
+            ]
+
+            self.absolute_pos_embed = nn.Parameter(
+                torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1])
+            )
+            trunc_normal_(self.absolute_pos_embed, std=0.02)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        # stochastic depth
+        dpr = [
+            x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))
+        ]  # stochastic depth decay rule
+
+        # build layers
+        self.layers = nn.ModuleList()
+        # prepare downsample list
+        downsamplelist = [PatchMerging for i in range(self.num_layers)]
+        downsamplelist[-1] = None
+        num_features = [int(embed_dim * 2**i) for i in range(self.num_layers)]
+        if self.dilation:
+            downsamplelist[-2] = None
+            num_features[-1] = int(embed_dim * 2 ** (self.num_layers - 1)) // 2
+        for i_layer in range(self.num_layers):
+            layer = BasicLayer(
+                # dim=int(embed_dim * 2 ** i_layer),
+                dim=num_features[i_layer],
+                depth=depths[i_layer],
+                num_heads=num_heads[i_layer],
+                window_size=window_size,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[sum(depths[:i_layer]) : sum(depths[: i_layer + 1])],
+                norm_layer=norm_layer,
+                # downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
+                downsample=downsamplelist[i_layer],
+                use_checkpoint=use_checkpoint,
+            )
+            self.layers.append(layer)
+
+        # num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)]
+        self.num_features = num_features
+
+        # add a norm layer for each output
+        for i_layer in out_indices:
+            layer = norm_layer(num_features[i_layer])
+            layer_name = f"norm{i_layer}"
+            self.add_module(layer_name, layer)
+
+        self._freeze_stages()
+
+    def _freeze_stages(self):
+        if self.frozen_stages >= 0:
+            self.patch_embed.eval()
+            for param in self.patch_embed.parameters():
+                param.requires_grad = False
+
+        if self.frozen_stages >= 1 and self.ape:
+            self.absolute_pos_embed.requires_grad = False
+
+        if self.frozen_stages >= 2:
+            self.pos_drop.eval()
+            for i in range(0, self.frozen_stages - 1):
+                m = self.layers[i]
+                m.eval()
+                for param in m.parameters():
+                    param.requires_grad = False
+
+    # def init_weights(self, pretrained=None):
+    #     """Initialize the weights in backbone.
+    #     Args:
+    #         pretrained (str, optional): Path to pre-trained weights.
+    #             Defaults to None.
+    #     """
+
+    #     def _init_weights(m):
+    #         if isinstance(m, nn.Linear):
+    #             trunc_normal_(m.weight, std=.02)
+    #             if isinstance(m, nn.Linear) and m.bias is not None:
+    #                 nn.init.constant_(m.bias, 0)
+    #         elif isinstance(m, nn.LayerNorm):
+    #             nn.init.constant_(m.bias, 0)
+    #             nn.init.constant_(m.weight, 1.0)
+
+    #     if isinstance(pretrained, str):
+    #         self.apply(_init_weights)
+    #         logger = get_root_logger()
+    #         load_checkpoint(self, pretrained, strict=False, logger=logger)
+    #     elif pretrained is None:
+    #         self.apply(_init_weights)
+    #     else:
+    #         raise TypeError('pretrained must be a str or None')
+
+    def forward_raw(self, x):
+        """Forward function."""
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+        if self.ape:
+            # interpolate the position embedding to the corresponding size
+            absolute_pos_embed = F.interpolate(
+                self.absolute_pos_embed, size=(Wh, Ww), mode="bicubic"
+            )
+            x = (x + absolute_pos_embed).flatten(2).transpose(1, 2)  # B Wh*Ww C
+        else:
+            x = x.flatten(2).transpose(1, 2)
+        x = self.pos_drop(x)
+
+        outs = []
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+            # import ipdb; ipdb.set_trace()
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f"norm{i}")
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+        # in:
+        #   torch.Size([2, 3, 1024, 1024])
+        # outs:
+        #   [torch.Size([2, 192, 256, 256]), torch.Size([2, 384, 128, 128]), \
+        #       torch.Size([2, 768, 64, 64]), torch.Size([2, 1536, 32, 32])]
+        return tuple(outs)
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+
+        """Forward function."""
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+        if self.ape:
+            # interpolate the position embedding to the corresponding size
+            absolute_pos_embed = F.interpolate(
+                self.absolute_pos_embed, size=(Wh, Ww), mode="bicubic"
+            )
+            x = (x + absolute_pos_embed).flatten(2).transpose(1, 2)  # B Wh*Ww C
+        else:
+            x = x.flatten(2).transpose(1, 2)
+        x = self.pos_drop(x)
+
+        outs = []
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f"norm{i}")
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+        # in:
+        #   torch.Size([2, 3, 1024, 1024])
+        # out:
+        #   [torch.Size([2, 192, 256, 256]), torch.Size([2, 384, 128, 128]), \
+        #       torch.Size([2, 768, 64, 64]), torch.Size([2, 1536, 32, 32])]
+
+        # collect for nesttensors
+        outs_dict = {}
+        for idx, out_i in enumerate(outs):
+            m = tensor_list.mask
+            assert m is not None
+            mask = F.interpolate(m[None].float(), size=out_i.shape[-2:]).to(torch.bool)[0]
+            outs_dict[idx] = NestedTensor(out_i, mask)
+
+        return outs_dict
+
+    def train(self, mode=True):
+        """Convert the model into training mode while keep layers freezed."""
+        super(SwinTransformer, self).train(mode)
+        self._freeze_stages()
+
+
+def build_swin_transformer(modelname, pretrain_img_size, **kw):
+    assert modelname in [
+        "swin_T_224_1k",
+        "swin_B_224_22k",
+        "swin_B_384_22k",
+        "swin_L_224_22k",
+        "swin_L_384_22k",
+    ]
+
+    model_para_dict = {
+        "swin_T_224_1k": dict(
+            embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], window_size=7
+        ),
+        "swin_B_224_22k": dict(
+            embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=7
+        ),
+        "swin_B_384_22k": dict(
+            embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=12
+        ),
+        "swin_L_224_22k": dict(
+            embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=7
+        ),
+        "swin_L_384_22k": dict(
+            embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=12
+        ),
+    }
+    kw_cgf = model_para_dict[modelname]
+    kw_cgf.update(kw)
+    model = SwinTransformer(pretrain_img_size=pretrain_img_size, **kw_cgf)
+    return model
+
+
+if __name__ == "__main__":
+    model = build_swin_transformer("swin_L_384_22k", 384, dilation=True)
+    x = torch.rand(2, 3, 1024, 1024)
+    y = model.forward_raw(x)
+    import ipdb
+
+    ipdb.set_trace()
+    x = torch.rand(2, 3, 384, 384)
+    y = model.forward_raw(x)

groundingdino/models/GroundingDINO/bertwarper.py

+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint as checkpoint
+from torch import Tensor, nn
+from torchvision.ops.boxes import nms
+from transformers import BertConfig, BertModel, BertPreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPoolingAndCrossAttentions
+
+
+class BertModelWarper(nn.Module):
+    def __init__(self, bert_model):
+        super().__init__()
+        # self.bert = bert_modelc
+
+        self.config = bert_model.config
+        self.embeddings = bert_model.embeddings
+        self.encoder = bert_model.encoder
+        self.pooler = bert_model.pooler
+
+        self.get_extended_attention_mask = bert_model.get_extended_attention_mask
+        self.invert_attention_mask = bert_model.invert_attention_mask
+        self.get_head_mask = bert_model.get_head_mask
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = (
+            output_attentions if output_attentions is not None else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if self.config.is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        # past_key_values_length
+        past_key_values_length = (
+            past_key_values[0][0].shape[2] if past_key_values is not None else 0
+        )
+
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                ((batch_size, seq_length + past_key_values_length)), device=device
+            )
+        if token_type_ids is None:
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(
+            attention_mask, input_shape, device
+        )
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.is_decoder and encoder_hidden_states is not None:
+            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+        # if os.environ.get('IPDB_SHILONG_DEBUG', None) == 'INFO':
+        #     import ipdb; ipdb.set_trace()
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+
+        embedding_output = self.embeddings(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            inputs_embeds=inputs_embeds,
+            past_key_values_length=past_key_values_length,
+        )
+
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+class TextEncoderShell(nn.Module):
+    def __init__(self, text_encoder):
+        super().__init__()
+        self.text_encoder = text_encoder
+        self.config = self.text_encoder.config
+
+    def forward(self, **kw):
+        # feed into text encoder
+        return self.text_encoder(**kw)
+
+
+def generate_masks_with_special_tokens(tokenized, special_tokens_list, tokenizer):
+    """Generate attention mask between each pair of special tokens
+    Args:
+        input_ids (torch.Tensor): input ids. Shape: [bs, num_token]
+        special_tokens_mask (list): special tokens mask.
+    Returns:
+        torch.Tensor: attention mask between each special tokens.
+    """
+    input_ids = tokenized["input_ids"]
+    bs, num_token = input_ids.shape
+    # special_tokens_mask: bs, num_token. 1 for special tokens. 0 for normal tokens
+    special_tokens_mask = torch.zeros((bs, num_token), device=input_ids.device).bool()
+    for special_token in special_tokens_list:
+        special_tokens_mask |= input_ids == special_token
+
+    # idxs: each row is a list of indices of special tokens
+    idxs = torch.nonzero(special_tokens_mask)
+
+    # generate attention mask and positional ids
+    attention_mask = (
+        torch.eye(num_token, device=input_ids.device).bool().unsqueeze(0).repeat(bs, 1, 1)
+    )
+    position_ids = torch.zeros((bs, num_token), device=input_ids.device)
+    previous_col = 0
+    for i in range(idxs.shape[0]):
+        row, col = idxs[i]
+        if (col == 0) or (col == num_token - 1):
+            attention_mask[row, col, col] = True
+            position_ids[row, col] = 0
+        else:
+            attention_mask[row, previous_col + 1 : col + 1, previous_col + 1 : col + 1] = True
+            position_ids[row, previous_col + 1 : col + 1] = torch.arange(
+                0, col - previous_col, device=input_ids.device
+            )
+
+        previous_col = col
+
+    # # padding mask
+    # padding_mask = tokenized['attention_mask']
+    # attention_mask = attention_mask & padding_mask.unsqueeze(1).bool() & padding_mask.unsqueeze(2).bool()
+
+    return attention_mask, position_ids.to(torch.long)
+
+
+def generate_masks_with_special_tokens_and_transfer_map(tokenized, special_tokens_list, tokenizer):
+    """Generate attention mask between each pair of special tokens
+    Args:
+        input_ids (torch.Tensor): input ids. Shape: [bs, num_token]
+        special_tokens_mask (list): special tokens mask.
+    Returns:
+        torch.Tensor: attention mask between each special tokens.
+    """
+    input_ids = tokenized["input_ids"]
+    bs, num_token = input_ids.shape
+    # special_tokens_mask: bs, num_token. 1 for special tokens. 0 for normal tokens
+    special_tokens_mask = torch.zeros((bs, num_token), device=input_ids.device).bool()
+    for special_token in special_tokens_list:
+        special_tokens_mask |= input_ids == special_token
+
+    # idxs: each row is a list of indices of special tokens
+    idxs = torch.nonzero(special_tokens_mask)
+
+    # generate attention mask and positional ids
+    attention_mask = (
+        torch.eye(num_token, device=input_ids.device).bool().unsqueeze(0).repeat(bs, 1, 1)
+    )
+    position_ids = torch.zeros((bs, num_token), device=input_ids.device)
+    cate_to_token_mask_list = [[] for _ in range(bs)]
+    previous_col = 0
+    for i in range(idxs.shape[0]):
+        row, col = idxs[i]
+        if (col == 0) or (col == num_token - 1):
+            attention_mask[row, col, col] = True
+            position_ids[row, col] = 0
+        else:
+            attention_mask[row, previous_col + 1 : col + 1, previous_col + 1 : col + 1] = True
+            position_ids[row, previous_col + 1 : col + 1] = torch.arange(
+                0, col - previous_col, device=input_ids.device
+            )
+            c2t_maski = torch.zeros((num_token), device=input_ids.device).bool()
+            c2t_maski[previous_col + 1 : col] = True
+            cate_to_token_mask_list[row].append(c2t_maski)
+        previous_col = col
+
+    cate_to_token_mask_list = [
+        torch.stack(cate_to_token_mask_listi, dim=0)
+        for cate_to_token_mask_listi in cate_to_token_mask_list
+    ]
+
+    # # padding mask
+    # padding_mask = tokenized['attention_mask']
+    # attention_mask = attention_mask & padding_mask.unsqueeze(1).bool() & padding_mask.unsqueeze(2).bool()
+
+    return attention_mask, position_ids.to(torch.long), cate_to_token_mask_list

groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn.h

+/*!
+**************************************************************************************************
+* Deformable DETR
+* Copyright (c) 2020 SenseTime. All Rights Reserved.
+* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+**************************************************************************************************
+* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
+**************************************************************************************************
+*/
+
+#pragma once
+
+#include "ms_deform_attn_cpu.h"
+
+#ifdef WITH_CUDA
+#include "ms_deform_attn_cuda.h"
+#endif
+
+namespace groundingdino {
+
+at::Tensor
+ms_deform_attn_forward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const int im2col_step)
+{
+    if (value.type().is_cuda())
+    {
+#ifdef WITH_CUDA
+        return ms_deform_attn_cuda_forward(
+            value, spatial_shapes, level_start_index, sampling_loc, attn_weight, im2col_step);
+#else
+        AT_ERROR("Not compiled with GPU support");
+#endif
+    }
+    AT_ERROR("Not implemented on the CPU");
+}
+
+std::vector<at::Tensor>
+ms_deform_attn_backward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const at::Tensor &grad_output,
+    const int im2col_step)
+{
+    if (value.type().is_cuda())
+    {
+#ifdef WITH_CUDA
+        return ms_deform_attn_cuda_backward(
+            value, spatial_shapes, level_start_index, sampling_loc, attn_weight, grad_output, im2col_step);
+#else
+        AT_ERROR("Not compiled with GPU support");
+#endif
+    }
+    AT_ERROR("Not implemented on the CPU");
+}
+
+} // namespace groundingdino
\ No newline at end of file

groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn_cpu.cpp

+/*!
+**************************************************************************************************
+* Deformable DETR
+* Copyright (c) 2020 SenseTime. All Rights Reserved.
+* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+**************************************************************************************************
+* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
+**************************************************************************************************
+*/
+
+#include <vector>
+
+#include <ATen/ATen.h>
+#include <ATen/cuda/CUDAContext.h>
+
+namespace groundingdino {
+
+at::Tensor
+ms_deform_attn_cpu_forward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const int im2col_step)
+{
+    AT_ERROR("Not implement on cpu");
+}
+
+std::vector<at::Tensor>
+ms_deform_attn_cpu_backward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const at::Tensor &grad_output,
+    const int im2col_step)
+{
+    AT_ERROR("Not implement on cpu");
+}
+
+} // namespace groundingdino

groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn_cpu.h

+/*!
+**************************************************************************************************
+* Deformable DETR
+* Copyright (c) 2020 SenseTime. All Rights Reserved.
+* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+**************************************************************************************************
+* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
+**************************************************************************************************
+*/
+
+#pragma once
+#include <torch/extension.h>
+
+namespace groundingdino {
+
+at::Tensor
+ms_deform_attn_cpu_forward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const int im2col_step);
+
+std::vector<at::Tensor>
+ms_deform_attn_cpu_backward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const at::Tensor &grad_output,
+    const int im2col_step);
+
+} // namespace groundingdino

groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn_cpu_hip.cpp

+// !!! This is a file automatically generated by hipify!!!
+#include <ATen/dtk_macros.h>
+/*!
+**************************************************************************************************
+* Deformable DETR
+* Copyright (c) 2020 SenseTime. All Rights Reserved.
+* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+**************************************************************************************************
+* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
+**************************************************************************************************
+*/
+
+#include <vector>
+
+#include <ATen/ATen.h>
+#include <ATen/hip/HIPContext.h>
+
+namespace groundingdino {
+
+at::Tensor
+ms_deform_attn_cpu_forward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const int im2col_step)
+{
+    AT_ERROR("Not implement on cpu");
+}
+
+std::vector<at::Tensor>
+ms_deform_attn_cpu_backward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const at::Tensor &grad_output,
+    const int im2col_step)
+{
+    AT_ERROR("Not implement on cpu");
+}
+
+} // namespace groundingdino

groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn_cuda.cu

+/*!
+**************************************************************************************************
+* Deformable DETR
+* Copyright (c) 2020 SenseTime. All Rights Reserved.
+* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+**************************************************************************************************
+* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
+**************************************************************************************************
+*/
+
+#include <vector>
+#include "ms_deform_im2col_cuda.cuh"
+
+#include <ATen/ATen.h>
+#include <ATen/cuda/CUDAContext.h>
+#include <cuda.h>
+#include <cuda_runtime.h>
+
+namespace groundingdino {
+
+at::Tensor ms_deform_attn_cuda_forward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const int im2col_step)
+{
+    AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
+    AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
+    AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
+    AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
+    AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
+
+    AT_ASSERTM(value.type().is_cuda(), "value must be a CUDA tensor");
+    AT_ASSERTM(spatial_shapes.type().is_cuda(), "spatial_shapes must be a CUDA tensor");
+    AT_ASSERTM(level_start_index.type().is_cuda(), "level_start_index must be a CUDA tensor");
+    AT_ASSERTM(sampling_loc.type().is_cuda(), "sampling_loc must be a CUDA tensor");
+    AT_ASSERTM(attn_weight.type().is_cuda(), "attn_weight must be a CUDA tensor");
+
+    const int batch = value.size(0);
+    const int spatial_size = value.size(1);
+    const int num_heads = value.size(2);
+    const int channels = value.size(3);
+
+    const int num_levels = spatial_shapes.size(0);
+
+    const int num_query = sampling_loc.size(1);
+    const int num_point = sampling_loc.size(4);
+
+    const int im2col_step_ = std::min(batch, im2col_step);
+
+    AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
+    
+    auto output = at::zeros({batch, num_query, num_heads, channels}, value.options());
+
+    const int batch_n = im2col_step_;
+    auto output_n = output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
+    auto per_value_size = spatial_size * num_heads * channels;
+    auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
+    auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
+    for (int n = 0; n < batch/im2col_step_; ++n)
+    {
+        auto columns = output_n.select(0, n);
+        AT_DISPATCH_FLOATING_TYPES(value.type(), "ms_deform_attn_forward_cuda", ([&] {
+            ms_deformable_im2col_cuda(at::cuda::getCurrentCUDAStream(),
+                value.data<scalar_t>() + n * im2col_step_ * per_value_size,
+                spatial_shapes.data<int64_t>(),
+                level_start_index.data<int64_t>(),
+                sampling_loc.data<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
+                attn_weight.data<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
+                batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
+                columns.data<scalar_t>());
+
+        }));
+    }
+
+    output = output.view({batch, num_query, num_heads*channels});
+
+    return output;
+}
+
+std::vector<at::Tensor> ms_deform_attn_cuda_backward(
+    const at::Tensor &value, 
+    const at::Tensor &spatial_shapes,
+    const at::Tensor &level_start_index,
+    const at::Tensor &sampling_loc,
+    const at::Tensor &attn_weight,
+    const at::Tensor &grad_output,
+    const int im2col_step)
+{
+
+    AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
+    AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
+    AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
+    AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
+    AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
+    AT_ASSERTM(grad_output.is_contiguous(), "grad_output tensor has to be contiguous");
+
+    AT_ASSERTM(value.type().is_cuda(), "value must be a CUDA tensor");
+    AT_ASSERTM(spatial_shapes.type().is_cuda(), "spatial_shapes must be a CUDA tensor");
+    AT_ASSERTM(level_start_index.type().is_cuda(), "level_start_index must be a CUDA tensor");
+    AT_ASSERTM(sampling_loc.type().is_cuda(), "sampling_loc must be a CUDA tensor");
+    AT_ASSERTM(attn_weight.type().is_cuda(), "attn_weight must be a CUDA tensor");
+    AT_ASSERTM(grad_output.type().is_cuda(), "grad_output must be a CUDA tensor");
+
+    const int batch = value.size(0);
+    const int spatial_size = value.size(1);
+    const int num_heads = value.size(2);
+    const int channels = value.size(3);
+
+    const int num_levels = spatial_shapes.size(0);
+
+    const int num_query = sampling_loc.size(1);
+    const int num_point = sampling_loc.size(4);
+
+    const int im2col_step_ = std::min(batch, im2col_step);
+
+    AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
+
+    auto grad_value = at::zeros_like(value);
+    auto grad_sampling_loc = at::zeros_like(sampling_loc);
+    auto grad_attn_weight = at::zeros_like(attn_weight);
+
+    const int batch_n = im2col_step_;
+    auto per_value_size = spatial_size * num_heads * channels;
+    auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
+    auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
+    auto grad_output_n = grad_output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
+    
+    for (int n = 0; n < batch/im2col_step_; ++n)
+    {
+        auto grad_output_g = grad_output_n.select(0, n);
+        AT_DISPATCH_FLOATING_TYPES(value.type(), "ms_deform_attn_backward_cuda", ([&] {
+            ms_deformable_col2im_cuda(at::cuda::getCurrentCUDAStream(),
+                                    grad_output_g.data<scalar_t>(),
+                                    value.data<scalar_t>() + n * im2col_step_ * per_value_size,
+                                    spatial_shapes.data<int64_t>(),
+                                    level_start_index.data<int64_t>(),
+                                    sampling_loc.data<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
+                                    attn_weight.data<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
+                                    batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
+                                    grad_value.data<scalar_t>() +  n * im2col_step_ * per_value_size,
+                                    grad_sampling_loc.data<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
+                                    grad_attn_weight.data<scalar_t>() + n * im2col_step_ * per_attn_weight_size);
+
+        }));
+    }
+
+    return {
+        grad_value, grad_sampling_loc, grad_attn_weight
+    };
+}
+
+} // namespace groundingdino
\ No newline at end of file