search

jingjingrenabc / dlformer

13 stars 0 forks source link

Retraining model is not possible #3

Open bsmietanka opened 2 years ago

bsmietanka commented 2 years ago

I tried retraining model for other sequence, but some custom hardcoded vgg model checkpoint is required that is not provided by the authors.

https://github.com/JingjingRenabc/dlformer/blob/1b15e46e97b30e1b52026fd972598f064e3c4bb1/dlformer/modules/losses/lpips_vi.py#L88

MatthewChang commented 1 year ago

Same. This does not seem reproducible as of now.

MatthewChang commented 1 year ago

I reached out to the authors and they confirmed that this should be the vgg16 checkpoint from here https://pytorch.org/vision/0.8/_modules/torchvision/models/vgg.html.