Question about repo and dataset

[ds22058]

Great work!
I met problems when I runned the render.py and render_val_dataset.py. It could not fint pretrained/stereo_layers/stereo_layers.yaml. Besides, the pretrained/download.sh was empty and could you please explain the detailed information about the dataset including the path and usage etc.
Looking forward to your rely .Thanks in advance

[palsol]

Hello!
You can find the pre-trained model right here in the repository:
https://github.com/SamsungLabs/MLI/tree/main/pretrained/model8_layers
https://github.com/SamsungLabs/MLI/tree/main/pretrained/model4_layers

You can find info and the small sample data of our dataset here:
https://samsunglabs.github.io/StereoLayers/sword/
small sample 1.6G

[ds22058]

Thank you for your explanantion! I have another question now. When I run render.py, the result is 'KeyError: 'dataloaders'' . I guess it is because there is no dict named dataloaders in the pretrained/model4_layers/tblock4.yaml, so I copy the one in the tblock4_train.yaml and run again, the result is still the same as before. I check the code and find a line 'dataloader = get_dataloader_from_params (config['dataloaders'], 'render') ' in render.py, but in the defination of function get_dataloader_from_params it writes 'dataloader_config = params['dataloaders']'. I think the first code should be modified to 'dataloader = get_dataloader_from_params (config, 'render') ' and I try it, it turns out to be

20it [00:00, 475.59it/s]

Process finished with exit code 139 (interrupted by signal 11: SIGSEGV)

The above error also apears when I run the render_val_dataset.py with parameters
--config
pretrained/model8_layers/tblock8.yaml
--val-dataset
datasets/dataset_additional/partb_ipad
--iteration
400000
--output-path
outputs

I really want to run all the code to see the result of your Great work. Could you please help me with these questions?

[palsol]

Hello! We updated this repo, and now you can generate your own validation dataset and render videos with it.

[Sunineu]

@ds22058
I met the same problem ’KeyError: 'dataloaders' ‘as you when running render.py, and I wonder if you have resolved this issue.". If you have already solved this problem, could you tell me how did you do it. Thank you in advance!

[ds22058]

@Sunineu I check the code and find a line 'dataloader = get_dataloader_from_params (config['dataloaders'], 'render') ' in render.py, but in the defination of function get_dataloader_from_params it writes 'dataloader_config = params['dataloaders']'. I think the first code should be modified to 'dataloader = get_dataloader_from_params (config, 'render') ' . The author has already corrected it.

[Sunineu]

@ds22058
Thank you for your reply .I have tried many times but there is still some errors.I would like to know what configuration you used to run render.py.
And,i have Imitated you to modify first code to 'dataloader = get_dataloader_from_params (config, 'render') ' .Here is my config and error:
python bin/render.py --config /home/MLI-main/configs/tblock8_train.yaml --checkpoints-path pretrained/model8_layers/model.gen_00660000.pt --iteration 6600000 --output-path results
{'kornia_aug': 'RandomMixUp', 'p': 0.9, 'lambda_val': [0, 0.4]}
Traceback (most recent call last):
File "bin/render.py", line 69, in
main()
File "bin/render.py", line 51, in main
trainer, loaded_iteration = create_trainer_load_weights_from_config(config=config,
File "/home/MLI-main/bin/lib/trainers/utils.py", line 44, in create_trainer_load_weights_from_config
trainer = getattr(trainers, config['trainer'])(config, eval_mode=True, device=device)
File "/home/MLI-main/bin/lib/trainers/trainer_simpli.py", line 71, in init
self.augs_mix = construct_kornia_aug(
File "/home/MLI-main/bin/lib/utils/augmentations.py", line 57, in construct_kornia_aug
augs_list = construct_list_of_kornia_augs(params)
File "/home/MLI-main/bin/lib/utils/augmentations.py", line 31, in construct_list_of_kornia_augs
augs.append(kornia_aug(**kornia_params))
TypeError: 'NoneType' object is not callable