Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix Automatic Download of Pretrained Weights in DETR #17712

Merged
merged 11 commits into from
Jun 21, 2022
Merged

Fix Automatic Download of Pretrained Weights in DETR #17712

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
20f2a5e
added use_backbone_pretrained
AnugunjNaman Jun 15, 2022
ef54f2d
style fixes
AnugunjNaman Jun 15, 2022
6068303
update
AnugunjNaman Jun 15, 2022
98a5953
Merge branch 'huggingface:main' into fix15764
AnugunjNaman Jun 17, 2022
66eae21
Update detr.mdx
AnugunjNaman Jun 17, 2022
6998672
Update detr.mdx
AnugunjNaman Jun 17, 2022
25b19b2
Update detr.mdx
AnugunjNaman Jun 21, 2022
d07dce1
Merge branch 'huggingface:main' into fix15764
AnugunjNaman Jun 21, 2022
76623ce
update using doc py
AnugunjNaman Jun 21, 2022
a3fa784
Update detr.mdx
AnugunjNaman Jun 21, 2022
2cd0e13
Update src/transformers/models/detr/configuration_detr.py
AnugunjNaman Jun 21, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions src/transformers/models/detr/configuration_detr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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`):
AnugunjNaman marked this conversation as resolved.
Show resolved Hide resolved
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):
Expand Down Expand Up @@ -147,6 +149,7 @@ def __init__(
auxiliary_loss=False,
position_embedding_type="sine",
backbone="resnet50",
use_pretrained_backbone=True,
dilation=False,
class_cost=1,
bbox_cost=5,
Expand Down Expand Up @@ -180,6 +183,7 @@ def __init__(
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
Expand Down
17 changes: 14 additions & 3 deletions src/transformers/models/detr/modeling_detr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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 = {}
Expand All @@ -335,7 +335,9 @@ def __init__(self, name: str, dilation: bool):

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)
Expand Down Expand Up @@ -1177,7 +1179,7 @@ def __init__(self, config: DetrConfig):
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)

Expand Down Expand Up @@ -1234,6 +1236,9 @@ def forward(
>>> image = Image.open(requests.get(url, stream=True).raw)

>>> feature_extractor = DetrFeatureExtractor.from_pretrained("facebook/detr-resnet-50")
>>> # model use pretrained backbone weights by default
>>> # to prevent this set use_pretrained_backbone = False
>>> # model = DetrModel.from_pretrained("facebook/detr-resnet-50", use_pretrained_backbone = False)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this makes sense, cause it will load pre-trained weights anyway.

A better code example (in my opinion) would be:

# randomly initialize a DETR model with pre-trained ResNet weights
config = DetrConfig()
model = DetrModel(config)

# randomly initialize a DETR model (with randomly initialized ResNet)
config = DetrConfig(use_pretrained_backbone=False)
model = DetrModel(config)

Copy link
Contributor Author

@AnugunjNaman AnugunjNaman Jun 15, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@NielsRogge. Done. Works like you asked.

>>> model = DetrModel.from_pretrained("facebook/detr-resnet-50")
>>> inputs = feature_extractor(images=image, return_tensors="pt")
>>> outputs = model(**inputs)
Expand Down Expand Up @@ -1392,6 +1397,9 @@ def forward(
>>> image = Image.open(requests.get(url, stream=True).raw)

>>> feature_extractor = DetrFeatureExtractor.from_pretrained("facebook/detr-resnet-50")
>>> # model use pretrained backbone weights by default
>>> # to prevent this set use_pretrained_backbone = False
>>> # model = DetrModel.from_pretrained("facebook/detr-resnet-50", use_pretrained_backbone = False)
>>> model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

>>> inputs = feature_extractor(images=image, return_tensors="pt")
Expand Down Expand Up @@ -1548,6 +1556,9 @@ def forward(
>>> image = Image.open(requests.get(url, stream=True).raw)

>>> feature_extractor = DetrFeatureExtractor.from_pretrained("facebook/detr-resnet-50-panoptic")
>>> # model use pretrained backbone weights by default
>>> # to prevent this set use_pretrained_backbone = False
>>> # model = DetrModel.from_pretrained("facebook/detr-resnet-50", use_pretrained_backbone = False)
>>> model = DetrForSegmentation.from_pretrained("facebook/detr-resnet-50-panoptic")

>>> inputs = feature_extractor(images=image, return_tensors="pt")
Expand Down