Add OWL-ViT model for zero-shot object detection (#17938)

* add owlvit model skeleton

* add class and box predictor heads

* convert modified flax clip to pytorch

* fix box and class predictors

* add OwlViTImageTextEmbedder

* convert class and box head checkpoints

* convert image text embedder checkpoints

* add object detection head

* fix bugs

* update conversion script

* update conversion script

* fix q,v,k,out weight conversion conversion

* add owlvit object detection output

* fix bug in image embedder

* fix bugs in text embedder

* fix positional embeddings

* fix bug in inference mode vision pooling

* update docs, init tokenizer and processor files

* support batch processing

* add OwlViTProcessor

* remove merge conflicts

* readd owlvit imports

* fix bug in OwlViTProcessor imports

* fix bugs in processor

* update docs

* fix bugs in processor

* update owlvit docs

* add OwlViTFeatureExtractor

* style changes, add postprocess method to feature extractor

* add feature extractor and processor tests

* add object detection tests

* update conversion script

* update config paths

* update config paths

* fix configuration paths and bugs

* fix bugs in OwlViT tests

* add import checks to processor

* fix docs and minor issues

* fix docs and minor issues

* fix bugs and issues

* fix bugs and issues

* fix bugs and issues

* fix bugs and issues

* update docs and examples

* fix bugs and issues

* update conversion script, fix positional embeddings

* process 2D input ids, update tests

* fix style and quality issues

* update docs

* update docs and imports

* update OWL-ViT index.md

* fix bug in OwlViT feature ext tests

* fix code examples, return_dict by default

* return_dict by default

* minor fixes, add tests to processor

* small fixes

* add output_attentions arg to main model

* fix bugs

* remove output_hidden_states arg from main model

* update self.config variables

* add option to return last_hidden_states

* fix bug in config variables

* fix copied from statements

* fix small issues and bugs

* fix bugs

* fix bugs, support greyscale images

* run fixup

* update repo name

* merge OwlViTImageTextEmbedder with obj detection head

* fix merge conflict

* fix merge conflict

* make fixup

* fix bugs

* fix bugs

* add additional processor test

src/transformers/models/owlvit/processing_owlvit.py

+# coding=utf-8
+# Copyright 2022 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Image/Text processor class for OWL-ViT
+"""
+from typing import List
+
+import numpy as np
+
+from transformers import is_flax_available, is_tf_available, is_torch_available
+
+from ...processing_utils import ProcessorMixin
+from ...tokenization_utils_base import BatchEncoding
+
+
+class OwlViTProcessor(ProcessorMixin):
+    r"""
+    Constructs an OWL-ViT processor which wraps [`OwlViTFeatureExtractor`] and [`CLIPTokenizer`]/[`CLIPTokenizerFast`]
+    into a single processor that interits both the feature extractor and tokenizer functionalities. See the
+    [`~OwlViTProcessor.__call__`] and [`~OwlViTProcessor.decode`] for more information.
+
+    Args:
+        feature_extractor ([`OwlViTFeatureExtractor`]):
+            The feature extractor is a required input.
+        tokenizer ([`CLIPTokenizer`, `CLIPTokenizerFast`]):
+            The tokenizer is a required input.
+    """
+    feature_extractor_class = "OwlViTFeatureExtractor"
+    tokenizer_class = ("CLIPTokenizer", "CLIPTokenizerFast")
+
+    def __init__(self, feature_extractor, tokenizer):
+        super().__init__(feature_extractor, tokenizer)
+
+    def __call__(self, text=None, images=None, padding="max_length", return_tensors="np", **kwargs):
+        """
+        Main method to prepare for the model one or several text(s) and image(s). This method forwards the `text` and
+        `kwargs` arguments to CLIPTokenizerFast's [`~CLIPTokenizerFast.__call__`] if `text` is not `None` to encode:
+        the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to
+        CLIPFeatureExtractor's [`~CLIPFeatureExtractor.__call__`] if `images` is not `None`. Please refer to the
+        doctsring of the above two methods for more information.
+
+        Args:
+            text (`str`, `List[str]`, `List[List[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`,
+            `List[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. In case of a NumPy array/PyTorch tensor, each image should be of shape (C, H, W), where C is a
+                number of channels, H and W are image height and width.
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+                - `'tf'`: Return TensorFlow `tf.constant` objects.
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+                - `'jax'`: Return JAX `jnp.ndarray` objects.
+        Returns:
+            [`BatchEncoding`]: A [`BatchEncoding`] with the following fields:
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+        """
+
+        if text is None and images is None:
+            raise ValueError("You have to specify at least one text or image. Both cannot be none.")
+
+        if text is not None:
+            if isinstance(text, str) or (isinstance(text, List) and not isinstance(text[0], List)):
+                encodings = [self.tokenizer(text, padding=padding, return_tensors=return_tensors, **kwargs)]
+
+            elif isinstance(text, List) and isinstance(text[0], List):
+                encodings = []
+
+                # Maximum number of queries across batch
+                max_num_queries = max([len(t) for t in text])
+
+                # Pad all batch samples to max number of text queries
+                for t in text:
+                    if len(t) != max_num_queries:
+                        t = t + [" "] * (max_num_queries - len(t))
+
+                    encoding = self.tokenizer(t, padding=padding, return_tensors=return_tensors, **kwargs)
+                    encodings.append(encoding)
+            else:
+                raise TypeError("Input text should be a string, a list of strings or a nested list of strings")
+
+            if return_tensors == "np":
+                input_ids = np.concatenate([encoding["input_ids"] for encoding in encodings], axis=0)
+                attention_mask = np.concatenate([encoding["attention_mask"] for encoding in encodings], axis=0)
+
+            elif return_tensors == "jax" and is_flax_available():
+                import jax.numpy as jnp
+
+                input_ids = jnp.concatenate([encoding["input_ids"] for encoding in encodings], axis=0)
+                attention_mask = jnp.concatenate([encoding["attention_mask"] for encoding in encodings], axis=0)
+
+            elif return_tensors == "pt" and is_torch_available():
+                import torch
+
+                input_ids = torch.cat([encoding["input_ids"] for encoding in encodings], dim=0)
+                attention_mask = torch.cat([encoding["attention_mask"] for encoding in encodings], dim=0)
+
+            elif return_tensors == "tf" and is_tf_available():
+                import tensorflow as tf
+
+                input_ids = tf.stack([encoding["input_ids"] for encoding in encodings], axis=0)
+                attention_mask = tf.stack([encoding["attention_mask"] for encoding in encodings], axis=0)
+
+            else:
+                raise ValueError("Target return tensor type could not be returned")
+
+            encoding = BatchEncoding()
+            encoding["input_ids"] = input_ids
+            encoding["attention_mask"] = attention_mask
+
+        if images is not None:
+            image_features = self.feature_extractor(images, return_tensors=return_tensors, **kwargs)
+
+        if text is not None and images is not None:
+            encoding["pixel_values"] = image_features.pixel_values
+            return encoding
+        elif text is not None:
+            return encoding
+        else:
+            return BatchEncoding(data=dict(**image_features), tensor_type=return_tensors)
+
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to CLIPTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to CLIPTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)

src/transformers/utils/dummy_pt_objects.py

@@ -3459,6 +3459,44 @@ class OPTPreTrainedModel(metaclass=DummyObject):
         requires_backends(self, ["torch"])
 
 
+OWLVIT_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class OwlViTForObjectDetection(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class OwlViTModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class OwlViTPreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class OwlViTTextModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class OwlViTVisionModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 class PegasusForCausalLM(metaclass=DummyObject):
     _backends = ["torch"]
 

src/transformers/utils/dummy_vision_objects.py

@@ -122,6 +122,13 @@ class MobileViTFeatureExtractor(metaclass=DummyObject):
         requires_backends(self, ["vision"])
 
 
+class OwlViTFeatureExtractor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
 class PerceiverFeatureExtractor(metaclass=DummyObject):
     _backends = ["vision"]
 

tests/models/owlvit/test_feature_extraction_owlvit.py

+# coding=utf-8
+# Copyright 2022 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import unittest
+
+import numpy as np
+
+from transformers.testing_utils import require_torch, require_vision
+from transformers.utils import is_torch_available, is_vision_available
+
+from ...test_feature_extraction_common import FeatureExtractionSavingTestMixin, prepare_image_inputs
+
+
+if is_torch_available():
+    import torch
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import OwlViTFeatureExtractor
+
+
+class OwlViTFeatureExtractionTester(unittest.TestCase):
+    def __init__(
+        self,
+        parent,
+        batch_size=7,
+        num_channels=3,
+        image_size=18,
+        min_resolution=30,
+        max_resolution=400,
+        do_resize=True,
+        size=20,
+        do_center_crop=True,
+        crop_size=18,
+        do_normalize=True,
+        image_mean=[0.48145466, 0.4578275, 0.40821073],
+        image_std=[0.26862954, 0.26130258, 0.27577711],
+        do_convert_rgb=True,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.min_resolution = min_resolution
+        self.max_resolution = max_resolution
+        self.do_resize = do_resize
+        self.size = size
+        self.do_center_crop = do_center_crop
+        self.crop_size = crop_size
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean
+        self.image_std = image_std
+        self.do_convert_rgb = do_convert_rgb
+
+    def prepare_feat_extract_dict(self):
+        return {
+            "do_resize": self.do_resize,
+            "size": self.size,
+            "do_center_crop": self.do_center_crop,
+            "crop_size": self.crop_size,
+            "do_normalize": self.do_normalize,
+            "image_mean": self.image_mean,
+            "image_std": self.image_std,
+            "do_convert_rgb": self.do_convert_rgb,
+        }
+
+
+@require_torch
+@require_vision
+class OwlViTFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCase):
+
+    feature_extraction_class = OwlViTFeatureExtractor if is_vision_available() else None
+
+    def setUp(self):
+        self.feature_extract_tester = OwlViTFeatureExtractionTester(self)
+
+    @property
+    def feat_extract_dict(self):
+        return self.feature_extract_tester.prepare_feat_extract_dict()
+
+    def test_feat_extract_properties(self):
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        self.assertTrue(hasattr(feature_extractor, "do_resize"))
+        self.assertTrue(hasattr(feature_extractor, "size"))
+        self.assertTrue(hasattr(feature_extractor, "do_center_crop"))
+        self.assertTrue(hasattr(feature_extractor, "center_crop"))
+        self.assertTrue(hasattr(feature_extractor, "do_normalize"))
+        self.assertTrue(hasattr(feature_extractor, "image_mean"))
+        self.assertTrue(hasattr(feature_extractor, "image_std"))
+        self.assertTrue(hasattr(feature_extractor, "do_convert_rgb"))
+
+    def test_call_pil(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PIL images
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
+
+        for image in image_inputs:
+            self.assertIsInstance(image, Image.Image)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )
+
+    def test_call_numpy(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random numpy tensors
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, numpify=True)
+        for image in image_inputs:
+            self.assertIsInstance(image, np.ndarray)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )
+
+    def test_call_pytorch(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PyTorch tensors
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, torchify=True)
+        for image in image_inputs:
+            self.assertIsInstance(image, torch.Tensor)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size,
+                self.feature_extract_tester.crop_size,
+            ),
+        )

tests/models/owlvit/test_processor_owlvit.py

+# Copyright 2022 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+import shutil
+import tempfile
+import unittest
+
+import numpy as np
+import pytest
+
+from transformers import CLIPTokenizer, CLIPTokenizerFast
+from transformers.models.clip.tokenization_clip import VOCAB_FILES_NAMES
+from transformers.testing_utils import require_vision
+from transformers.utils import FEATURE_EXTRACTOR_NAME, is_vision_available
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import OwlViTFeatureExtractor, OwlViTProcessor
+
+
+@require_vision
+class OwlViTProcessorTest(unittest.TestCase):
+    def setUp(self):
+        self.tmpdirname = tempfile.mkdtemp()
+
+        # fmt: off
+        vocab = ["", "l", "o", "w", "e", "r", "s", "t", "i", "d", "n", "lo", "l</w>", "w</w>", "r</w>", "t</w>", "low</w>", "er</w>", "lowest</w>", "newer</w>", "wider", "<unk>", "<|startoftext|>", "<|endoftext|>"]
+        # fmt: on
+        vocab_tokens = dict(zip(vocab, range(len(vocab))))
+        merges = ["#version: 0.2", "l o", "lo w</w>", "e r</w>", ""]
+        self.special_tokens_map = {"unk_token": "<unk>"}
+
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["merges_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as fp:
+            fp.write(json.dumps(vocab_tokens) + "\n")
+        with open(self.merges_file, "w", encoding="utf-8") as fp:
+            fp.write("\n".join(merges))
+
+        feature_extractor_map = {
+            "do_resize": True,
+            "size": 20,
+            "do_center_crop": True,
+            "crop_size": 18,
+            "do_normalize": True,
+            "image_mean": [0.48145466, 0.4578275, 0.40821073],
+            "image_std": [0.26862954, 0.26130258, 0.27577711],
+        }
+        self.feature_extractor_file = os.path.join(self.tmpdirname, FEATURE_EXTRACTOR_NAME)
+        with open(self.feature_extractor_file, "w", encoding="utf-8") as fp:
+            json.dump(feature_extractor_map, fp)
+
+    def get_tokenizer(self, **kwargs):
+        return CLIPTokenizer.from_pretrained(self.tmpdirname, pad_token="!", **kwargs)
+
+    def get_rust_tokenizer(self, **kwargs):
+        return CLIPTokenizerFast.from_pretrained(self.tmpdirname, pad_token="!", **kwargs)
+
+    def get_feature_extractor(self, **kwargs):
+        return OwlViTFeatureExtractor.from_pretrained(self.tmpdirname, **kwargs)
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdirname)
+
+    def prepare_image_inputs(self):
+        """This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
+        or a list of PyTorch tensors if one specifies torchify=True.
+        """
+
+        image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
+
+        image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
+
+        return image_inputs
+
+    def test_save_load_pretrained_default(self):
+        tokenizer_slow = self.get_tokenizer()
+        tokenizer_fast = self.get_rust_tokenizer()
+        feature_extractor = self.get_feature_extractor()
+
+        processor_slow = OwlViTProcessor(tokenizer=tokenizer_slow, feature_extractor=feature_extractor)
+        processor_slow.save_pretrained(self.tmpdirname)
+        processor_slow = OwlViTProcessor.from_pretrained(self.tmpdirname, use_fast=False)
+
+        processor_fast = OwlViTProcessor(tokenizer=tokenizer_fast, feature_extractor=feature_extractor)
+        processor_fast.save_pretrained(self.tmpdirname)
+        processor_fast = OwlViTProcessor.from_pretrained(self.tmpdirname)
+
+        self.assertEqual(processor_slow.tokenizer.get_vocab(), tokenizer_slow.get_vocab())
+        self.assertEqual(processor_fast.tokenizer.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertEqual(tokenizer_slow.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertIsInstance(processor_slow.tokenizer, CLIPTokenizer)
+        self.assertIsInstance(processor_fast.tokenizer, CLIPTokenizerFast)
+
+        self.assertEqual(processor_slow.feature_extractor.to_json_string(), feature_extractor.to_json_string())
+        self.assertEqual(processor_fast.feature_extractor.to_json_string(), feature_extractor.to_json_string())
+        self.assertIsInstance(processor_slow.feature_extractor, OwlViTFeatureExtractor)
+        self.assertIsInstance(processor_fast.feature_extractor, OwlViTFeatureExtractor)
+
+    def test_save_load_pretrained_additional_features(self):
+        processor = OwlViTProcessor(tokenizer=self.get_tokenizer(), feature_extractor=self.get_feature_extractor())
+        processor.save_pretrained(self.tmpdirname)
+
+        tokenizer_add_kwargs = self.get_tokenizer(bos_token="(BOS)", eos_token="(EOS)")
+        feature_extractor_add_kwargs = self.get_feature_extractor(do_normalize=False)
+
+        processor = OwlViTProcessor.from_pretrained(
+            self.tmpdirname, bos_token="(BOS)", eos_token="(EOS)", do_normalize=False
+        )
+
+        self.assertEqual(processor.tokenizer.get_vocab(), tokenizer_add_kwargs.get_vocab())
+        self.assertIsInstance(processor.tokenizer, CLIPTokenizerFast)
+
+        self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
+        self.assertIsInstance(processor.feature_extractor, OwlViTFeatureExtractor)
+
+    def test_feature_extractor(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = OwlViTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        image_input = self.prepare_image_inputs()
+
+        input_feat_extract = feature_extractor(image_input, return_tensors="np")
+        input_processor = processor(images=image_input, return_tensors="np")
+
+        for key in input_feat_extract.keys():
+            self.assertAlmostEqual(input_feat_extract[key].sum(), input_processor[key].sum(), delta=1e-2)
+
+    def test_tokenizer(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = OwlViTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        input_str = "lower newer"
+
+        encoded_processor = processor(text=input_str, return_tensors="np")
+
+        encoded_tok = tokenizer(input_str, return_tensors="np")
+
+        for key in encoded_tok.keys():
+            self.assertListEqual(encoded_tok[key][0].tolist(), encoded_processor[key][0].tolist())
+
+    def test_processor(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = OwlViTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        input_str = "lower newer"
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(text=input_str, images=image_input)
+
+        self.assertListEqual(list(inputs.keys()), ["input_ids", "attention_mask", "pixel_values"])
+
+        # test if it raises when no input is passed
+        with pytest.raises(ValueError):
+            processor()
+
+    def test_processor_with_text_list(self):
+        model_name = "google/owlvit-base-patch32"
+        processor = OwlViTProcessor.from_pretrained(model_name)
+
+        input_text = ["cat", "nasa badge"]
+        inputs = processor(text=input_text)
+
+        seq_length = 16
+        self.assertListEqual(list(inputs.keys()), ["input_ids", "attention_mask"])
+        self.assertEqual(inputs["input_ids"].shape, (2, seq_length))
+
+        # test if it raises when no input is passed
+        with pytest.raises(ValueError):
+            processor()
+
+    def test_processor_with_nested_text_list(self):
+        model_name = "google/owlvit-base-patch32"
+        processor = OwlViTProcessor.from_pretrained(model_name)
+
+        input_texts = [["cat", "nasa badge"], ["person"]]
+        inputs = processor(text=input_texts)
+
+        seq_length = 16
+        batch_size = len(input_texts)
+        num_max_text_queries = max([len(texts) for texts in input_texts])
+
+        self.assertListEqual(list(inputs.keys()), ["input_ids", "attention_mask"])
+        self.assertEqual(inputs["input_ids"].shape, (batch_size * num_max_text_queries, seq_length))
+
+        # test if it raises when no input is passed
+        with pytest.raises(ValueError):
+            processor()
+
+    def test_processor_case(self):
+        model_name = "google/owlvit-base-patch32"
+        processor = OwlViTProcessor.from_pretrained(model_name)
+
+        input_texts = ["cat", "nasa badge"]
+        inputs = processor(text=input_texts)
+
+        seq_length = 16
+        input_ids = inputs["input_ids"]
+        predicted_ids = [
+            [49406, 2368, 49407, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+            [49406, 6841, 11301, 49407, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+        ]
+
+        self.assertListEqual(list(inputs.keys()), ["input_ids", "attention_mask"])
+        self.assertEqual(inputs["input_ids"].shape, (2, seq_length))
+        self.assertListEqual(list(input_ids[0]), predicted_ids[0])
+        self.assertListEqual(list(input_ids[1]), predicted_ids[1])
+
+    def test_tokenizer_decode(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = OwlViTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        predicted_ids = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]
+
+        decoded_processor = processor.batch_decode(predicted_ids)
+        decoded_tok = tokenizer.batch_decode(predicted_ids)
+
+        self.assertListEqual(decoded_tok, decoded_processor)

utils/check_config_docstrings.py

@@ -41,6 +41,7 @@ _re_checkpoint = re.compile("\[(.+?)\]\((https://huggingface\.co/.+?)\)")
 
 CONFIG_CLASSES_TO_IGNORE_FOR_DOCSTRING_CHECKPOINT_CHECK = {
     "CLIPConfig",
+    "OwlViTConfig",
     "GroupViTConfig",
     "DecisionTransformerConfig",
     "EncoderDecoderConfig",

utils/check_repo.py

@@ -166,6 +166,9 @@ IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
     "LukeForEntityPairClassification",
     "LukeForEntitySpanClassification",
     "OpenAIGPTDoubleHeadsModel",
+    "OwlViTTextModel",
+    "OwlViTVisionModel",
+    "OwlViTForObjectDetection",
     "RagModel",
     "RagSequenceForGeneration",
     "RagTokenForGeneration",