search

OPHoperHPO / image-background-remove-tool

✂️ Automated high-quality background removal framework for an image using neural networks. ✂️
https://carve.photos
Apache License 2.0
1.39k stars 272 forks source link

Issue and possible future deterrents #62

Closed cramik closed 2 years ago

cramik commented 3 years ago

At the moment I'm getting this issue with my dataset of dogs that I'm removing the backgrounds from (base) cramik@DESKTOP-G88PL8G:/mnt/c/dog/image-background-remove-tool$ python3 main.py -i /mnt/c/dogwitbg -o /mnt/c/dogwitoutbg -m u2net -prep bbd-fastrcnn -postp rtb-bnb Remove Background: 23%|############1 | 452/2001 [2:11:51<6:52:10, 15.97s/image]Traceback (most recent call last): File "main.py", line 152, in <module> cli() File "main.py", line 148, in cli process(input_path, output_path, model_name, preprocessing_method_name, postprocessing_method_name) File "main.py", line 113, in process image = model.process_image(file_path, preprocessing_method, postprocessing_method) File "/mnt/c/dog/image-background-remove-tool/libs/networks.py", line 94, in process_image image, org_image = self.__load_image__(data) # Load image File "/mnt/c/dog/image-background-remove-tool/libs/networks.py", line 153, in __load_image__ image = self.__ndrarray2tensor__(image) # Convert image from numpy arr to tensor File "/mnt/c/dog/image-background-remove-tool/libs/networks.py", line 164, in __ndrarray2tensor__ if image.shape[2] == 1: IndexError: tuple index out of range

I'm not sure what this is caused by but to prevent it or even troubleshoot it, I think it could be useful to either have an option to show the file name that is being accessed or even add an argument to delete files that have been processed or add them to a database so the user can restart the process without losing progress.

cramik commented 3 years ago

at least one of the issues was that it was treating folders as files btw, I'm still running the dataset though so I'll have to see if another one comes up

OPHoperHPO commented 2 years ago

Filename display. Currently not feasible. In the current version, image processing is carried out in batches, so the appeal goes directly to several photos at the same time. In view of which, in the case of a large array of images, progress display will break due to the large amount of information displayed in the terminal.

Issues with folders. Fixed in the latest release.