This repository has been archived by the owner on Mar 12, 2024. It is now read-only.
Add torchscript support for hub detection models #51
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
fmassa
commented
Jun 4, 2020
models/backbone.py
Outdated
| xs = self.body(tensor_list.tensors) | ||
| out = OrderedDict() | ||
| out : Dict[str, NestedTensor] = {} |
There was a problem hiding this comment.
This can be a BC-breaking change for some versions of Python, as we are changing OrderedDict to dict.
If this is a problem, we can change it to be a list.
fmassa
commented
Jun 4, 2020
| return out | ||
|
|
||
| @torch.jit.unused | ||
| def _set_aux_loss(self, outputs_class, outputs_coord): |
There was a problem hiding this comment.
aux_loss is not supported for exporting to torchscript, but this should be fine as we can always remove it for inference.
There was a problem hiding this comment.
does aux_loss only ever have pred_logits and pred_boxes as keys ? Could use a namedtuple maybe ?
There was a problem hiding this comment.
This might vary depending on the task. But I guess we could leave some fields of the namedtuple as None?
fmassa
commented
Jun 4, 2020
|
|
||
| def decompose(self): | ||
| return self.tensors, self.mask | ||
|
|
||
| @classmethod |
There was a problem hiding this comment.
classmethod doesn't seem to be supported, so had to move this as a separate function (with a few extra changes to make it compatible with torchscript)
fmassa
commented
Jun 4, 2020
| @@ -64,15 +65,17 @@ def __init__(self, backbone: nn.Module, train_backbone: bool, num_channels: int, | |||
| if return_interm_layers: | |||
| return_layers = {"layer1": "0", "layer2": "1", "layer3": "2", "layer4": "3"} | |||
| else: | |||
| return_layers = {'layer4': 0} | |||
| return_layers = {'layer4': "0"} | |||
There was a problem hiding this comment.
This oversight took me a while to figure out, as the error messages were pretty cryptic
Error message from torchscript
Traceback (most recent call last):
File "test_all.py", line 64, in test_model_script
scripted_model = torch.jit.script(model)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 1340, in script
return torch.jit._recursive.create_script_module(obj, torch.jit._recursive.infer_methods_to_compile)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 313, in create_script_module
return create_script_module_impl(nn_module, concrete_type, stubs_fn)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 367, in create_script_module_impl
script_module = torch.jit.RecursiveScriptModule._construct(cpp_module, init_fn)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 1713, in _construct
init_fn(script_module)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 348, in init_fn
scripted = create_script_module_impl(orig_value, sub_concrete_type, infer_methods_to_compile)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 367, in create_script_module_impl
script_module = torch.jit.RecursiveScriptModule._construct(cpp_module, init_fn)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 1713, in _construct
init_fn(script_module)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 348, in init_fn
scripted = create_script_module_impl(orig_value, sub_concrete_type, infer_methods_to_compile)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 367, in create_script_module_impl
script_module = torch.jit.RecursiveScriptModule._construct(cpp_module, init_fn)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 1713, in _construct
init_fn(script_module)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 348, in init_fn
scripted = create_script_module_impl(orig_value, sub_concrete_type, infer_methods_to_compile)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 367, in create_script_module_impl
script_module = torch.jit.RecursiveScriptModule._construct(cpp_module, init_fn)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 1713, in _construct
init_fn(script_module)
File "/Users/fmassa/anaconda3/lib/python3.6/site-packages/torch/jit/_recursive.py", line 333, in init_fn
cpp_module.setattr(name, orig_value)
RuntimeError: Unable to cast Python instance to C++ type (compile in debug mode for details)cc @eellison if this error message could be improved it would be great.
alcinos
approved these changes
Jun 4, 2020
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds support for torchscript export to the detection models in DETR torchhub.
It requires a recent-enough PyTorch (nightly from today worked for me), and doesn't work with PyTorch 1.5.0.