Update huggingface model

Update huggingface model


Update README.md


Update README.md


Update README.md


Update huggingface model


Update huggingface model

classification/huggingface/22k_model/internimage_l_22k_384/preprocessor_config.json

+{
+  "crop_size": 384,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "size": 384
+}

classification/huggingface/22k_model/internimage_xl_22k_384/README.md

+---
+license: mit
+pipeline_tag: image-classification
+library_name: transformers
+tags:
+  - internimage
+  - custom_code
+datasets:
+  - ILSVRC/imagenet-1k
+---
+
+# InternImage Model Card
+
+## Introduction
+
+InternImage is an advanced vision foundation model developed by researchers from Shanghai AI Laboratory, Tsinghua University, and other institutions. Unlike models based on Transformers, InternImage employs DCNv3 as its core operator. This approach equips the model with dynamic and effective receptive fields required for downstream tasks like object detection and segmentation, while enabling adaptive spatial aggregation.
+
+<div style="text-align: center;"> <img src="https://github.com/OpenGVLab/InternImage/raw/master/docs/figs/arch.png" style="width:60%;" /> </div>
+
+## Performance
+
+- InternImage achieved an impressive Top-1 accuracy of 90.1% on the ImageNet benchmark dataset using only publicly available data for image classification. Apart from two undisclosed models trained with additional datasets by Google and Microsoft, InternImage is the only open-source model that achieves a Top-1 accuracy of over 90.0%, and it is also the largest model in scale worldwide.
+- InternImage outperformed all other models worldwide on the COCO object detection benchmark dataset with a remarkable mAP of 65.5, making it the only model that surpasses 65 mAP in the world.
+- InternImage also demonstrated world's best performance on 16 other important visual benchmark datasets, covering a wide range of tasks such as classification, detection, and segmentation, making it the top-performing model across multiple domains.
+
+## Released Models
+
+### Open‑Source Visual Pretrained Models
+
+|                                       huggingface name                                        |  model  name   |       pretrain       | resolution | #param |
+| :-------------------------------------------------------------------------------------------: | :------------: | :------------------: | :--------: | :----: |
+|        [internimage_l_22k_384](https://huggingface.co/OpenGVLab/internimage_l_22k_384)        | InternImage-L  |        IN-22K        |  384x384   |  223M  |
+|       [internimage_xl_22k_384](https://huggingface.co/OpenGVLab/internimage_xl_22k_384)       | InternImage-XL |        IN-22K        |  384x384   |  335M  |
+| [internimage_h_jointto22k_384](https://huggingface.co/OpenGVLab/internimage_h_jointto22k_384) | InternImage-H  | Joint 427M -> IN-22K |  384x384   | 1.08B  |
+| [internimage_g_jointto22k_384](https://huggingface.co/OpenGVLab/internimage_g_jointto22k_384) | InternImage-G  | Joint 427M -> IN-22K |  384x384   |   3B   |
+
+### ImageNet-1K Image Classification
+
+|                                     huggingface name                                      |   model name   |       pretrain       | resolution | acc@1 | #param | FLOPs |
+| :---------------------------------------------------------------------------------------: | :------------: | :------------------: | :--------: | :---: | :----: | :---: |
+|       [internimage_t_1k_224](https://huggingface.co/OpenGVLab/internimage_t_1k_224)       | InternImage-T  |        IN-1K         |  224x224   | 83.5  |  30M   |  5G   |
+|       [internimage_s_1k_224](https://huggingface.co/OpenGVLab/internimage_s_1k_224)       | InternImage-S  |        IN-1K         |  224x224   | 84.2  |  50M   |  8G   |
+|       [internimage_b_1k_224](https://huggingface.co/OpenGVLab/internimage_b_1k_224)       | InternImage-B  |        IN-1K         |  224x224   | 84.9  |  97M   |  16G  |
+|  [internimage_l_22kto1k_384](https://huggingface.co/OpenGVLab/internimage_l_22kto1k_384)  | InternImage-L  |        IN-22K        |  384x384   | 87.7  |  223M  | 108G  |
+| [internimage_xl_22kto1k_384](https://huggingface.co/OpenGVLab/internimage_xl_22kto1k_384) | InternImage-XL |        IN-22K        |  384x384   | 88.0  |  335M  | 163G  |
+|  [internimage_h_22kto1k_640](https://huggingface.co/OpenGVLab/internimage_h_22kto1k_640)  | InternImage-H  | Joint 427M -> IN-22K |  640x640   | 89.6  | 1.08B  | 1478G |
+|  [internimage_g_22kto1k_512](https://huggingface.co/OpenGVLab/internimage_g_22kto1k_512)  | InternImage-G  | Joint 427M -> IN-22K |  512x512   | 90.1  |   3B   | 2700G |
+
+## DCNv3 CUDA Kernel Installation
+
+If you do not install the CUDA version of DCNv3, InternImage will automatically fall back to a PyTorch implementation. However, the CUDA implementation can significantly reduce GPU memory usage and improve inference efficiency.
+
+**Installation Tutorial:**
+
+1. Open your terminal and run:
+
+   ```bash
+   git clone https://github.com/OpenGVLab/InternImage.git
+   cd InternImage/classification/ops_dcnv3
+   ```
+
+2. Make sure you have an available GPU for compilation, then run:
+
+   ```bash
+   sh make.sh
+   ```
+
+This will compile the CUDA version of DCNv3. Once installed, InternImage will automatically leverage the optimized CUDA implementation for better performance.
+
+## Usage with Transformers
+
+Below are two usage examples for InternImage with the Transformers framework:
+
+### Example 1: Using InternImage as an Image Backbone
+
+```python
+import torch
+from PIL import Image
+from transformers import AutoModel, CLIPImageProcessor
+
+# Replace 'model_name' with the appropriate model identifier
+model_name = "OpenGVLab/internimage_t_1k_224"  # example model
+
+# Prepare the image
+image_path = 'img.png'
+image_processor = CLIPImageProcessor.from_pretrained(model_name)
+image = Image.open(image_path)
+image = image_processor(images=image, return_tensors='pt').pixel_values
+print('image shape:', image.shape)
+
+# Load the model as a backbone
+model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
+# 'hidden_states' contains the outputs from the 4 stages of the InternImage backbone
+hidden_states = model(image).hidden_states
+```
+
+### Example 2: Using InternImage for Image Classification
+
+```python
+import torch
+from PIL import Image
+from transformers import AutoModelForImageClassification, CLIPImageProcessor
+
+# Replace 'model_name' with the appropriate model identifier
+model_name = "OpenGVLab/internimage_t_1k_224"  # example model
+
+# Prepare the image
+image_path = 'img.png'
+image_processor = CLIPImageProcessor.from_pretrained(model_name)
+image = Image.open(image_path)
+image = image_processor(images=image, return_tensors='pt').pixel_values
+print('image shape:', image.shape)
+
+# Load the model as an image classifier
+model = AutoModelForImageClassification.from_pretrained(model_name, trust_remote_code=True)
+logits = model(image).logits
+label = torch.argmax(logits, dim=1)
+print("Predicted label:", label.item())
+```
+
+## Citation
+
+If this work is helpful for your research, please consider citing the following BibTeX entry.
+
+```Bibtex
+@inproceedings{wang2023internimage,
+  title={Internimage: Exploring large-scale vision foundation models with deformable convolutions},
+  author={Wang, Wenhai and Dai, Jifeng and Chen, Zhe and Huang, Zhenhang and Li, Zhiqi and Zhu, Xizhou and Hu, Xiaowei and Lu, Tong and Lu, Lewei and Li, Hongsheng and others},
+  booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
+  pages={14408--14419},
+  year={2023}
+}
+```

classification/huggingface/22k_model/internimage_xl_22k_384/config.json

+{
+  "_name_or_path": "OpenGVLab/internimage_xl_22k_384",
+  "act_layer": "GELU",
+  "architectures": [
+    "InternImageModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_internimage.InternImageConfig",
+    "AutoModel": "modeling_internimage.InternImageModel",
+    "AutoModelForImageClassification": "modeling_internimage.InternImageModelForImageClassification"
+  },
+  "center_feature_scale": false,
+  "channels": 192,
+  "cls_scale": 1.5,
+  "core_op": "DCNv3",
+  "depths": [
+    5,
+    5,
+    24,
+    5
+  ],
+  "drop_path_rate": 0.0,
+  "drop_path_type": "linear",
+  "drop_rate": 0.0,
+  "dw_kernel_size": null,
+  "groups": [
+    12,
+    24,
+    48,
+    96
+  ],
+  "layer_scale": 1e-05,
+  "level2_post_norm": false,
+  "level2_post_norm_block_ids": null,
+  "mlp_ratio": 4.0,
+  "model_type": "internimage",
+  "norm_layer": "LN",
+  "num_classes": 21841,
+  "offset_scale": 2.0,
+  "post_norm": true,
+  "remove_center": false,
+  "res_post_norm": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.37.2",
+  "use_clip_projector": false,
+  "with_cp": false
+}

classification/huggingface/22k_model/internimage_xl_22k_384/configuration_internimage.py

+# --------------------------------------------------------
+# InternImage
+# Copyright (c) 2025 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+
+from transformers import PretrainedConfig
+
+
+class InternImageConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`~InternImageModel`].
+    It is used to instantiate an internimage model according to the specified arguments, defining the model
+    architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
+    the internimage [OpenGVLab/internimage](https://huggingface.co/OpenGVLab/internimage) architecture.
+
+    Configuration objects inherit from  [`PretrainedConfig`] and can be used
+    to control the model outputs. Read the documentation from  [`PretrainedConfig`]
+    for more information.
+
+    Args:
+        core_op (`str`, *optional*, defaults to `"DCNv3"`):
+            Core operation used in the InternImageModel.
+        depths (`tuple`, *optional*, defaults to `(4, 4, 18, 4)`):
+            Tuple specifying the depth of layers in the InternImageModel.
+        groups (`tuple`, *optional*, defaults to `(4, 8, 16, 32)`):
+            Tuple specifying the group of layers in the InternImageModel.
+        channels (`int`, *optional*, defaults to `64`):
+            Number of channels in the InternImageModel.
+        dw_kernel_size (`int`, *optional*, defaults to `None`):
+            Kernel size for depthwise convolutions.
+        layer_scale (`float`, *optional*, defaults to `None`):
+            Scale of the layers in the model.
+        offset_scale (`float`, *optional*, defaults to `1.0`):
+            Offset scale in the model.
+        mlp_ratio (`float`, *optional*, defaults to `4.0`):
+            Ratio of mlp layers in the InternImageModel.
+        post_norm (`bool`, *optional*, defaults to `False`):
+            Whether to use post normalization in the model.
+        level2_post_norm (`bool`, *optional*, defaults to `False`):
+            Whether to use level 2 post normalization.
+        level2_post_norm_block_ids (`list`, *optional*, defaults to `None`):
+            Specific block IDs for level 2 post normalization.
+        center_feature_scale (`bool`, *optional*, defaults to `False`):
+            Whether to apply center feature scaling.
+        use_clip_projector (`bool`, *optional*, defaults to `False`):
+            Whether to use CLIP projector.
+        remove_center (`bool`, *optional*, defaults to `False`):
+            Whether to remove center pixels in some operations.
+        num_classes (`int`, *optional*, defaults to `1000`):
+            Number of classes for the model output.
+        drop_rate (`float`, *optional*, defaults to `0.0`):
+            Dropout rate in the model.
+        drop_path_rate (`float`, *optional*, defaults to `0.0`):
+            Dropout path rate in the model.
+        drop_path_type (`str`, *optional*, defaults to `"linear"`):
+            Type of dropout path used in the model.
+        act_layer (`str`, *optional*, defaults to `"GELU"`):
+            Activation function used in the model.
+        norm_layer (`str`, *optional*, defaults to `"LN"`):
+            Normalization layer used in the model.
+        cls_scale (`float`, *optional*, defaults to `1.5`):
+            Scale of the classification layer in the model.
+        with_cp (`bool`, *optional*, defaults to `False`):
+            Whether to use checkpointing in the model.
+    """
+    model_type = 'internimage'
+
+    def __init__(
+            self,
+            core_op='DCNv3',
+            depths=(4, 4, 18, 4),
+            groups=(4, 8, 16, 32),
+            channels=64,
+            dw_kernel_size=None,
+            layer_scale=None,
+            offset_scale=1.0,
+            mlp_ratio=4.0,
+            post_norm=False,
+            res_post_norm=False,
+            level2_post_norm=False,
+            level2_post_norm_block_ids=None,
+            center_feature_scale=False,
+            use_clip_projector=False,
+            remove_center=False,
+            num_classes=1000,
+            drop_rate=0.0,
+            drop_path_rate=0.0,
+            drop_path_type='linear',
+            act_layer='GELU',
+            norm_layer='LN',
+            cls_scale=1.5,
+            with_cp=False,
+            **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        # Model configuration parameters
+        self.core_op = core_op
+        self.depths = depths
+        self.groups = groups
+        self.channels = channels
+        self.dw_kernel_size = dw_kernel_size
+        self.layer_scale = layer_scale
+        self.offset_scale = offset_scale
+        self.mlp_ratio = mlp_ratio
+        self.post_norm = post_norm
+        self.res_post_norm = res_post_norm
+        self.level2_post_norm = level2_post_norm
+        self.level2_post_norm_block_ids = level2_post_norm_block_ids
+        self.center_feature_scale = center_feature_scale
+        self.use_clip_projector = use_clip_projector
+        self.remove_center = remove_center
+        self.num_classes = num_classes
+        self.drop_rate = drop_rate
+        self.drop_path_rate = drop_path_rate
+        self.drop_path_type = drop_path_type
+        self.act_layer = act_layer
+        self.norm_layer = norm_layer
+        self.cls_scale = cls_scale
+        self.with_cp = with_cp

classification/huggingface/22k_model/internimage_xl_22k_384/preprocessor_config.json

+{
+  "crop_size": 384,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "size": 384
+}

classification/huggingface/README.md

+---
+license: mit
+pipeline_tag: image-classification
+library_name: transformers
+tags:
+  - internimage
+  - custom_code
+datasets:
+  - ILSVRC/imagenet-1k
+---
+
+# InternImage Model Card
+
+## Introduction
+
+InternImage is an advanced vision foundation model developed by researchers from Shanghai AI Laboratory, Tsinghua University, and other institutions. Unlike models based on Transformers, InternImage employs DCNv3 as its core operator. This approach equips the model with dynamic and effective receptive fields required for downstream tasks like object detection and segmentation, while enabling adaptive spatial aggregation.
+
+<div style="text-align: center;"> <img src="https://github.com/OpenGVLab/InternImage/raw/master/docs/figs/arch.png" style="width:60%;" /> </div>
+
+## Performance
+
+- InternImage achieved an impressive Top-1 accuracy of 90.1% on the ImageNet benchmark dataset using only publicly available data for image classification. Apart from two undisclosed models trained with additional datasets by Google and Microsoft, InternImage is the only open-source model that achieves a Top-1 accuracy of over 90.0%, and it is also the largest model in scale worldwide.
+- InternImage outperformed all other models worldwide on the COCO object detection benchmark dataset with a remarkable mAP of 65.5, making it the only model that surpasses 65 mAP in the world.
+- InternImage also demonstrated world's best performance on 16 other important visual benchmark datasets, covering a wide range of tasks such as classification, detection, and segmentation, making it the top-performing model across multiple domains.
+
+## Released Models
+
+### Open‑Source Visual Pretrained Models
+
+|                                       huggingface name                                        |  model  name   |       pretrain       | resolution | #param |
+| :-------------------------------------------------------------------------------------------: | :------------: | :------------------: | :--------: | :----: |
+|        [internimage_l_22k_384](https://huggingface.co/OpenGVLab/internimage_l_22k_384)        | InternImage-L  |        IN-22K        |  384x384   |  223M  |
+|       [internimage_xl_22k_384](https://huggingface.co/OpenGVLab/internimage_xl_22k_384)       | InternImage-XL |        IN-22K        |  384x384   |  335M  |
+| [internimage_h_jointto22k_384](https://huggingface.co/OpenGVLab/internimage_h_jointto22k_384) | InternImage-H  | Joint 427M -> IN-22K |  384x384   | 1.08B  |
+| [internimage_g_jointto22k_384](https://huggingface.co/OpenGVLab/internimage_g_jointto22k_384) | InternImage-G  | Joint 427M -> IN-22K |  384x384   |   3B   |
+
+### ImageNet-1K Image Classification
+
+|                                     huggingface name                                      |   model name   |       pretrain       | resolution | acc@1 | #param | FLOPs |
+| :---------------------------------------------------------------------------------------: | :------------: | :------------------: | :--------: | :---: | :----: | :---: |
+|       [internimage_t_1k_224](https://huggingface.co/OpenGVLab/internimage_t_1k_224)       | InternImage-T  |        IN-1K         |  224x224   | 83.5  |  30M   |  5G   |
+|       [internimage_s_1k_224](https://huggingface.co/OpenGVLab/internimage_s_1k_224)       | InternImage-S  |        IN-1K         |  224x224   | 84.2  |  50M   |  8G   |
+|       [internimage_b_1k_224](https://huggingface.co/OpenGVLab/internimage_b_1k_224)       | InternImage-B  |        IN-1K         |  224x224   | 84.9  |  97M   |  16G  |
+|  [internimage_l_22kto1k_384](https://huggingface.co/OpenGVLab/internimage_l_22kto1k_384)  | InternImage-L  |        IN-22K        |  384x384   | 87.7  |  223M  | 108G  |
+| [internimage_xl_22kto1k_384](https://huggingface.co/OpenGVLab/internimage_xl_22kto1k_384) | InternImage-XL |        IN-22K        |  384x384   | 88.0  |  335M  | 163G  |
+|  [internimage_h_22kto1k_640](https://huggingface.co/OpenGVLab/internimage_h_22kto1k_640)  | InternImage-H  | Joint 427M -> IN-22K |  640x640   | 89.6  | 1.08B  | 1478G |
+|  [internimage_g_22kto1k_512](https://huggingface.co/OpenGVLab/internimage_g_22kto1k_512)  | InternImage-G  | Joint 427M -> IN-22K |  512x512   | 90.1  |   3B   | 2700G |
+
+## DCNv3 CUDA Kernel Installation
+
+If you do not install the CUDA version of DCNv3, InternImage will automatically fall back to a PyTorch implementation. However, the CUDA implementation can significantly reduce GPU memory usage and improve inference efficiency.
+
+**Installation Tutorial:**
+
+1. Open your terminal and run:
+
+   ```bash
+   git clone https://github.com/OpenGVLab/InternImage.git
+   cd InternImage/classification/ops_dcnv3
+   ```
+
+2. Make sure you have an available GPU for compilation, then run:
+
+   ```bash
+   sh make.sh
+   ```
+
+This will compile the CUDA version of DCNv3. Once installed, InternImage will automatically leverage the optimized CUDA implementation for better performance.
+
+## Usage with Transformers
+
+Below are two usage examples for InternImage with the Transformers framework:
+
+### Example 1: Using InternImage as an Image Backbone
+
+```python
+import torch
+from PIL import Image
+from transformers import AutoModel, CLIPImageProcessor
+
+# Replace 'model_name' with the appropriate model identifier
+model_name = "OpenGVLab/internimage_t_1k_224"  # example model
+
+# Prepare the image
+image_path = 'img.png'
+image_processor = CLIPImageProcessor.from_pretrained(model_name)
+image = Image.open(image_path)
+image = image_processor(images=image, return_tensors='pt').pixel_values
+print('image shape:', image.shape)
+
+# Load the model as a backbone
+model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
+# 'hidden_states' contains the outputs from the 4 stages of the InternImage backbone
+hidden_states = model(image).hidden_states
+```
+
+### Example 2: Using InternImage for Image Classification
+
+```python
+import torch
+from PIL import Image
+from transformers import AutoModelForImageClassification, CLIPImageProcessor
+
+# Replace 'model_name' with the appropriate model identifier
+model_name = "OpenGVLab/internimage_t_1k_224"  # example model
+
+# Prepare the image
+image_path = 'img.png'
+image_processor = CLIPImageProcessor.from_pretrained(model_name)
+image = Image.open(image_path)
+image = image_processor(images=image, return_tensors='pt').pixel_values
+print('image shape:', image.shape)
+
+# Load the model as an image classifier
+model = AutoModelForImageClassification.from_pretrained(model_name, trust_remote_code=True)
+logits = model(image).logits
+label = torch.argmax(logits, dim=1)
+print("Predicted label:", label.item())
+```
+
+## Citation
+
+If this work is helpful for your research, please consider citing the following BibTeX entry.
+
+```Bibtex
+@inproceedings{wang2023internimage,
+  title={Internimage: Exploring large-scale vision foundation models with deformable convolutions},
+  author={Wang, Wenhai and Dai, Jifeng and Chen, Zhe and Huang, Zhenhang and Li, Zhiqi and Zhu, Xizhou and Hu, Xiaowei and Lu, Tong and Lu, Lewei and Li, Hongsheng and others},
+  booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
+  pages={14408--14419},
+  year={2023}
+}
+```

classification/huggingface/in1k_model/internimage_b_1k_224/preprocessor_config.json

+{
+  "crop_size": 224,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "size": 224
+}