Fix Automatic Download of Pretrained Weights in DETR (#17712)

* added use_backbone_pretrained

* style fixes

* update

* Update detr.mdx

* Update detr.mdx

* Update detr.mdx

* update using doc py

* Update detr.mdx

* Update src/transformers/models/detr/configuration_detr.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

docs/source/en/model_doc/detr.mdx

@@ -113,6 +113,28 @@ Tips:
 - The size of the images will determine the amount of memory being used, and will thus determine the `batch_size`.
   It is advised to use a batch size of 2 per GPU. See [this Github thread](https://github.com/facebookresearch/detr/issues/150) for more info.
 
+There are three ways to instantiate a DETR model (depending on what you prefer):
+  
+Option 1: Instantiate DETR with pre-trained weights for entire model
+```py
+>>> from transformers import DetrForObjectDetection
+
+>>> model = DetrForObjectDetection.from_pretrained("facebook/resnet-50")
+```
+
+Option 2: Instantiate DETR with randomly initialized weights for Transformer, but pre-trained weights for backbone
+```py
+>>> from transformers import DetrConfig, DetrForObjectDetection
+
+>>> config = DetrConfig()
+>>> model = DetrForObjectDetection(config)
+```
+Option 3: Instantiate DETR with randomly initialized weights for backbone + Transformer
+```py
+>>> config = DetrConfig(use_pretrained_backbone=False)
+>>> model = DetrForObjectDetection(config)
+```
+
 As a summary, consider the following table:
 
 | Task | Object detection | Instance segmentation | Panoptic segmentation |
@@ -166,4 +188,4 @@ mean Average Precision (mAP) and Panoptic Quality (PQ). The latter objects are i
 ## DetrForSegmentation
 
 [[autodoc]] DetrForSegmentation
-    - forward
\ No newline at end of file
+    - forward

src/transformers/models/detr/configuration_detr.py

@@ -82,6 +82,8 @@ class DetrConfig(PretrainedConfig):
             Name of convolutional backbone to use. Supports any convolutional backbone from the timm package. For a
             list of all available models, see [this
             page](https://rwightman.github.io/pytorch-image-models/#load-a-pretrained-model).
+        use_pretrained_backbone (`bool`, *optional*, defaults to `True`):
+            Whether to use pretrained weights for the backbone.
         dilation (`bool`, *optional*, defaults to `False`):
             Whether to replace stride with dilation in the last convolutional block (DC5).
         class_cost (`float`, *optional*, defaults to 1):
@@ -147,6 +149,7 @@ class DetrConfig(PretrainedConfig):
         auxiliary_loss=False,
         position_embedding_type="sine",
         backbone="resnet50",
+        use_pretrained_backbone=True,
         dilation=False,
         class_cost=1,
         bbox_cost=5,
@@ -180,6 +183,7 @@ class DetrConfig(PretrainedConfig):
         self.auxiliary_loss = auxiliary_loss
         self.position_embedding_type = position_embedding_type
         self.backbone = backbone
+        self.use_pretrained_backbone = use_pretrained_backbone
         self.dilation = dilation
         # Hungarian matcher
         self.class_cost = class_cost

src/transformers/models/detr/modeling_detr.py

@@ -326,7 +326,7 @@ class DetrTimmConvEncoder(nn.Module):
 
     """
 
-    def __init__(self, name: str, dilation: bool):
+    def __init__(self, name: str, dilation: bool, use_pretrained_backbone: bool):
         super().__init__()
 
         kwargs = {}
@@ -335,7 +335,9 @@ class DetrTimmConvEncoder(nn.Module):
 
         requires_backends(self, ["timm"])
 
-        backbone = create_model(name, pretrained=True, features_only=True, out_indices=(1, 2, 3, 4), **kwargs)
+        backbone = create_model(
+            name, pretrained=use_pretrained_backbone, features_only=True, out_indices=(1, 2, 3, 4), **kwargs
+        )
         # replace batch norm by frozen batch norm
         with torch.no_grad():
             replace_batch_norm(backbone)
@@ -1177,7 +1179,7 @@ class DetrModel(DetrPreTrainedModel):
         super().__init__(config)
 
         # Create backbone + positional encoding
-        backbone = DetrTimmConvEncoder(config.backbone, config.dilation)
+        backbone = DetrTimmConvEncoder(config.backbone, config.dilation, config.use_pretrained_backbone)
         position_embeddings = build_position_encoding(config)
         self.backbone = DetrConvModel(backbone, position_embeddings)