adamdad / dery Goto Github PK

When running 3 stage (Reassemby) , I meet the following obstacles:

KeyError: "ImageClassifier: 'DeRy is not in the models registry

Hi Authors, When I try to run get_rep.py, it raised the error: ModuleNotFoundError: No module named 'matplotlib.blocking_input'. It seems to be attributed to 'mmcls', so How should I do?

The version of my matplotlib is 3.7.1.

Thank you for your great job and it make sense to me.

I am curious about the role of NASWOT in DeRy. Generally, NASWOT score can be obtained from a randomly initialized networks, and can inflect the expressive ability of a neural network structure.

However, I am unsure whether the model in the code you provided

If the model is randomly initialized, I am curious whether you have tried loading pre-trained weights and comparing the results. Are there any drawbacks to using pre-trained weights that you have encountered in your experiments?"

Thanks for sharing the code and providing detailed instructions. However, even though I put imagenet data with the correct extension into the default directory, the code raised an issue:RuntimeError: ImageNet: Found 0 files in subfolders of: data/imagenet/val. Supported extensions are: .jpg,.jpeg,.png,.ppm,.bmp,.pgm,.tif Would you please help me sort out this or provide a runnable instruction?

I'm very interested in this work of yours and am trying to run the code. But when the command PYTHONPATH="$PWD" python simlarity/zeroshot_reassembly.py is executed at the end, an error is always reported. The reason for my analysis was that I could not get the best model in the prescribed rounds, but I increased the number of trials to a larger value and still could not find the best model. I wonder how the author views this issue?

Thanks for providing interesting works and publicly releasing the code implementation.

I tried to follow the instructions and run the code, but I encountered some obstacles and got stuck (in stage 1).

1. mmcv version: since the current default version is 1.7.0 (I then tried to downgrade to 1.5.0)
2. DeRy/blocksize/block_meta.py does not have MODEL2MODULES, and this leads to the import error in DeRy/blocksize/__init__.py.
3. DeRy/mmcls and mmls package are conflicted, and renaming the DeRy/mmcls to other names can address this issue
4. DeRy/mmcls/datasets/multi_label.py is missing, and I tried to remove all parts that imported multi_label but some errors are still shown.
5. What should I assign for $Config for stage 1 (Model Zoo Preparation)?

Has anyone (or authors) successfully launched the training based on the current version?

This is an interesting work, and I would appreciate it if the authors can address some errors in the code and provide the launchable code 🙂

Thank you.

This paper is really an amazing work that would influence alot in the whole realm!
Scenario1:
e.g. Reassemble with ResNet101, FasterRCNN backbone, and DeepLabv3 backbone, check if Acc@top1 of DeRy model could be better than ResNet101 on ImageNet?

Scenario2:
e.g. Reassemble with ResNet101 on dataset 1, on dataset 2, on dataset 3, check if Acc@top1 of DeRy model could be better than ResNet101 on dataset 1?

Hi Authors,

Thanks for your work.

When I tried to run get_rep.py, it seems we need to pass a parameter into “--config”. In your line 129 of file get_rep.py, you added a help saying "test config file path". However, I am sorry that I may not be able to get it.

For example, if we run: python get_rep.py --config ???   

Could you please provide a running script like the above one?

Thanks