Enhance Device and Precision Handling, and Improve Error Messages in DepthPro Model #35
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
|
hey @Amael, please review this. |
carlos-bg
reviewed
Oct 22, 2024
src/depth_pro/depth_pro.py
Outdated
| device: torch.device = torch.device("cpu"), | ||
| precision: torch.dtype = torch.float32, | ||
| device: torch.device = torch.device("cuda" if torch.cuda.is_available() else "cpu"), | ||
| precision: torch.dtype = torch.float16 if device.type == 'cuda' else torch.float32, |
There was a problem hiding this comment.
This use of device at this point doesn't seem to work for me.
I have to do this precision = torch.float16 if device.type == 'cuda' else torch.float32 in the body of the function
There was a problem hiding this comment.
When running depth-pro-run -i PATH_TO_MY_IMG_FILE, I also get a problem:
Traceback (most recent call last):
File "miniconda3/envs/depth-pro/bin/depth-pro-run", line 8, in <module>
sys.exit(run_main())
File "ml-depth-pro/src/depth_pro/cli/run.py", line 150, in main
run(parser.parse_args())
File "ml-depth-pro/src/depth_pro/cli/run.py", line 68, in run
prediction = model.infer(transform(image), f_px=f_px)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torchvision/transforms/transforms.py", line 95, in __call__
img = t(img)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
return forward_call(*args, **kwargs)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torchvision/transforms/transforms.py", line 198, in forward
return F.convert_image_dtype(image, self.dtype)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torchvision/transforms/functional.py", line 243, in convert_image_dtype
return F_t.convert_image_dtype(image, dtype)
File "miniconda3/envs/depth-pro/lib/python3.9/site-packages/torchvision/transforms/_functional_tensor.py", line 73, in convert_image_dtype
if torch.tensor(0, dtype=dtype).is_floating_point():
TypeError: tensor(): argument 'dtype' must be torch.dtype, not tuple
|
@carlos-bg i have made some changes please review this. |
|
@carlos-bg please review |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Pull Request Description
This pull request introduces several improvements to the DepthPro model code:
Device and Precision Handling: The model now dynamically selects the appropriate device (
cudaorcpu) based on availability. Additionally, handling for half precision (torch.half) has been implemented to enhance performance on compatible devices.Improved Error Messages: Enhanced error messages for loading model state dictionaries provide clearer feedback on any issues that arise during the loading process.
These enhancements aim to improve the usability and performance of the DepthPro model, making it more efficient and user-friendly.