Add TF ResNet model (#17427)

* Rought TF conversion outline

* Tidy up

* Fix padding differences between layers

* Add back embedder - whoops

* Match test file to main

* Match upstream test file

* Correctly pass and assign image_size parameter

Co-authored-by: Sayak Paul <[email protected]>

* Add in MainLayer

* Correctly name layer

* Tidy up AdaptivePooler

* Small tidy-up

More accurate type hints and remove whitespaces

* Change AdaptiveAvgPool

Use the AdaptiveAvgPool implementation by @Rocketknight1, which correctly pools if the output shape does not evenly divide by input shape c.f. https://github.com/huggingface/transformers/pull/17554/files/9e26607e22aa8d069c86b50196656012ff0ce62a#r900109509

Co-authored-by: From: matt <[email protected]>
Co-authored-by: Sayak Paul <[email protected]>

* Use updated AdaptiveAvgPool

Co-authored-by: matt <[email protected]>

* Make AdaptiveAvgPool compatible with CPU

* Remove image_size from configuration

* Fixup

* Tensorflow -> TensorFlow

* Fix pt references in tests

* Apply suggestions from code review - grammar and wording

Co-authored-by: NielsRogge <[email protected]>

Co-authored-by: NielsRogge <[email protected]>

* Add TFResNet to doc tests

* PR comments - GlobalAveragePooling and clearer comments

* Remove unused import

* Add in keepdims argument

* Add num_channels check

* grammar fix: by -> of

Co-authored-by: matt <[email protected]>

Co-authored-by: Matt <[email protected]>

* Remove transposes - keep NHWC throughout forward pass

* Fixup look sharp

* Add missing layer names

* Final tidy up - remove from_pt now weights on hub

Co-authored-by: Sayak Paul <[email protected]>
Co-authored-by: matt <[email protected]>
Co-authored-by: NielsRogge <[email protected]>
Co-authored-by: Matt <[email protected]>

Author: 4 people

docs/source/en/index.mdx

@@ -273,7 +273,7 @@ Flax), PyTorch, and/or TensorFlow.
 |          Reformer           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |           RegNet            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |           RemBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|           ResNet            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           ResNet            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |

docs/source/en/model_doc/resnet.mdx

@@ -31,7 +31,7 @@ The figure below illustrates the architecture of ResNet. Taken from the [origina
 
 <img width="600" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/resnet_architecture.png"/>
 
-This model was contributed by [Francesco](https://huggingface.co/Francesco). The original code can be found [here](https://github.com/KaimingHe/deep-residual-networks).
+This model was contributed by [Francesco](https://huggingface.co/Francesco). The TensorFlow version of this model was added by [amyeroberts](https://huggingface.co/amyeroberts). The original code can be found [here](https://github.com/KaimingHe/deep-residual-networks).
 
 ## ResNetConfig
 
@@ -47,4 +47,16 @@ This model was contributed by [Francesco](https://huggingface.co/Francesco). The
 ## ResNetForImageClassification
 
 [[autodoc]] ResNetForImageClassification
-    - forward
+    - forward
+
+
+## TFResNetModel
+
+[[autodoc]] TFResNetModel
+    - call
+
+
+## TFResNetForImageClassification
+
+[[autodoc]] TFResNetForImageClassification
+    - call

src/transformers/__init__.py

@@ -2380,6 +2380,14 @@
             "TFRemBertPreTrainedModel",
         ]
     )
+    _import_structure["models.resnet"].extend(
+        [
+            "TF_RESNET_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "TFResNetForImageClassification",
+            "TFResNetModel",
+            "TFResNetPreTrainedModel",
+        ]
+    )
     _import_structure["models.roberta"].extend(
         [
             "TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -4721,6 +4729,12 @@
             TFRemBertModel,
             TFRemBertPreTrainedModel,
         )
+        from .models.resnet import (
+            TF_RESNET_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFResNetForImageClassification,
+            TFResNetModel,
+            TFResNetPreTrainedModel,
+        )
         from .models.roberta import (
             TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFRobertaForCausalLM,

src/transformers/modeling_tf_outputs.py

@@ -62,7 +62,7 @@ class TFBaseModelOutputWithNoAttention(ModelOutput):
     """
 
     last_hidden_state: tf.Tensor = None
-    hidden_states: Optional[Tuple[tf.Tensor]] = None
+    hidden_states: Optional[Tuple[tf.Tensor, ...]] = None
 
 
 @dataclass
@@ -118,7 +118,7 @@ class TFBaseModelOutputWithPoolingAndNoAttention(ModelOutput):
 
     last_hidden_state: tf.Tensor = None
     pooler_output: tf.Tensor = None
-    hidden_states: Optional[Tuple[tf.Tensor]] = None
+    hidden_states: Optional[Tuple[tf.Tensor, ...]] = None
 
 
 @dataclass
@@ -886,4 +886,4 @@ class TFImageClassifierOutputWithNoAttention(ModelOutput):
 
     loss: Optional[tf.Tensor] = None
     logits: tf.Tensor = None
-    hidden_states: Optional[Tuple[tf.Tensor]] = None
+    hidden_states: Optional[Tuple[tf.Tensor, ...]] = None

src/transformers/models/auto/modeling_tf_auto.py

@@ -64,6 +64,7 @@
         ("pegasus", "TFPegasusModel"),
         ("regnet", "TFRegNetModel"),
         ("rembert", "TFRemBertModel"),
+        ("resnet", "TFResNetModel"),
         ("roberta", "TFRobertaModel"),
         ("roformer", "TFRoFormerModel"),
         ("speech_to_text", "TFSpeech2TextModel"),
@@ -175,6 +176,7 @@
         ("convnext", "TFConvNextForImageClassification"),
         ("data2vec-vision", "TFData2VecVisionForImageClassification"),
         ("regnet", "TFRegNetForImageClassification"),
+        ("resnet", "TFResNetForImageClassification"),
         ("swin", "TFSwinForImageClassification"),
         ("vit", "TFViTForImageClassification"),
     ]

src/transformers/models/resnet/__init__.py

@@ -18,7 +18,7 @@
 from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
-from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
 
 
 _import_structure = {
@@ -38,6 +38,19 @@
         "ResNetPreTrainedModel",
     ]
 
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_tf_resnet"] = [
+        "TF_RESNET_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "TFResNetForImageClassification",
+        "TFResNetModel",
+        "TFResNetPreTrainedModel",
+    ]
+
 
 if TYPE_CHECKING:
     from .configuration_resnet import RESNET_PRETRAINED_CONFIG_ARCHIVE_MAP, ResNetConfig, ResNetOnnxConfig
@@ -55,6 +68,19 @@
             ResNetPreTrainedModel,
         )
 
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_tf_resnet import (
+            TF_RESNET_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFResNetForImageClassification,
+            TFResNetModel,
+            TFResNetPreTrainedModel,
+        )
+
 
 else:
     import sys