search

zlai0 / MAST

MAST: A Memory-Augmented Self-supervised Tracker (CVPR 2020)
https://zlai0.github.io/MAST/
273 stars 32 forks source link

Where is the memory module in training stage? #6

Closed AlanIIE closed 3 years ago

AlanIIE commented 4 years ago

Very perfect work!

I admit that the test results on the provided pre-trained model are the same as the released ones. However, with the given codes, I cannot train a model with similar performance on the YouTube-VOS dataset on my own. My model just got Js=0.405 and Fs=0.481 with 30 epochs' training. What's the problem?

When I check the training model in main.py (Ln 184), I find it's only trained on pairwise data. Could you please release the code of the long and short term memory as told in the paper? Is it the reason why I cannot achieve a higher score?

zlai0 commented 4 years ago

That's weird. What's the versions of your dependencies (Pytorch etc.)?

AlanIIE commented 4 years ago

Thanks for your quick reply. I trained on the dependencies: python==3.7 CUDA=10.0, torch==1.2.0, torchvision==0.4.0, spatial-correlation-sampler==0.2.0

ruoqi77 commented 4 years ago

I have the same question too. Maybe we should change dataloader code to load more images as reference ?

zlai0 commented 4 years ago

Thanks for your quick reply. I trained on the dependencies: python==3.7 CUDA=10.0, torch==1.2.0, torchvision==0.4.0, spatial-correlation-sampler==0.2.0

You may need the same versions as specified in readme. There are some compatibility issue, e.g. I found pytorch 1.4 doesn't work.

DexiangHong commented 3 years ago

@zlai0 Thank you for sharing your excellent work! But I have the same problem with @AlanIIE. With the given code and default setting, I just got Js = 0.392 . Do I need add multi-frame training ?

DexiangHong commented 3 years ago

@zlai0 Thank you for sharing your excellent work! But I have the same problem with @AlanIIE. With the given code and default setting, I just got Js = 0.392 . Do I need add multi-frame training ?

pytorch 1.2 also does not work. I use pytorch 1.1.0 and spatial-correlation-sampler 0.0.8, successfully reproduce the results in the paper. That's so weird!!