search

ByungKwanLee / Causal-Unsupervised-Segmentation

Official PyTorch Implementation code for realizing the technical part of Causal Unsupervised Semantic sEgmentation (CAUSE) to improve performance of unsupervised semantic segmentation. (Under Review)
11 stars 1 forks source link

Data with no label #11

Closed AriaeiMehdi closed 4 months ago

AriaeiMehdi commented 4 months ago

Hi, I would like to know if it is possible to train CAUSE without labels available?

thanks Mehdi.

AkankshaP0102 commented 4 months ago

Hey @AriaeiMehdi did you get any luck in using the model without labels. Please can you let me know the results. Thank you in Advance.

AriaeiMehdi commented 4 months ago

Hey @AriaeiMehdi did you get any luck in using the model without labels. Please can you let me know the results. Thank you in Advance.

Hi, In my case I had some labels available so I didn't get to train without labels. As far as I understand, you can train the mediator script (train_mediator.py)with no labels. here is the part of the code that just loads the images and trains the network without labels: https://github.com/ByungKwanLee/Causal-Unsupervised-Segmentation/blob/45cf51029e2580f53bff618bf48598b974cc8482/train_mediator.py#L30-L32

but the author used labels in train_front_door.py and fine_tuning.py(tr and mlp) script to calculate linear loss. you may need to ignore linear loss in these scripts to train without labels.