search

suhwan-cho / TMO

[WACV 2023] Treating Motion as Option to Reduce Motion Dependency in Unsupervised Video Object Segmentation
MIT License
56 stars 3 forks source link

There is a problem in model training #10

Closed zp19990818 closed 1 year ago

zp19990818 commented 1 year ago

Hi,

my dataset is binary segment data and has been modified in util.py. It can be trained, and the loss shows that it is going down. But the output iou has not been quite normal. I have two datasets; one has stabilized at 1 in less than 20 rounds of iou, and the other iou has been 0. Has the initialization of the model created any problems?

0e71deccdec56e7715d25fef833a180

9af1b704ad42a3eea2b49eb9ae630ce

suhwan-cho commented 1 year ago

Hi,

Is training going well with the original datasets?

zp19990818 commented 1 year ago

Hi,

Is training going well with the original datasets?

Yes, they train normally on the original dataset. But my data is also normal and runnable, I think it might be the dataset distribution as well as the network initialization. I'll go ahead and debug it myself, thanks for your advice!

suhwan-cho commented 1 year ago

Great, no worries!

zp19990818 commented 1 year ago

Great, no worries!

I saw your suggestion in another discussion (https://github.com/suhwan-cho/TMO/issues/7) and tried to modify the object pixel values from [255,255,255] to [128,0,0]. It takes a while to wait, but the training works. Thanks for the suggestion, but I think there may indeed be some problem here, but I can't locate it.

Anyway, thanks again!