search

juntang-zhuang / ShelfNet

implementation for paper "ShelfNet for fast semantic segmentation"
MIT License
252 stars 41 forks source link

Tips for better training #20

Closed mive93 closed 3 years ago

mive93 commented 4 years ago

Dear @juntang-zhuang,

First of all, thank you for this repo. I am trying to use it to train shelfnet on the Mapillary Vistas Dataset (here you can find my fork). I have succeeded training she Real-Time version of Shelfnet, however the results are pretty bad even after 270000 epochs. The reached mIOU is 34,06%, however in this paper they say that they were able to achieve 49.2% on your model. I have already tried to contact the authors, but I got no response.
Therefore I wanted to ask you if you could give me some tips to improve the training to achieve better results.

Thank you in advance.

Best, Micaela

juntang-zhuang commented 4 years ago

Hi, sorry I have not experimented on Mapillary, too large for me. My suggestion would be first to check your dataloader and class definition, any tiny errors there could cause a big difference. Second is you can try a smaller learning rate or a different optimizer such as Adam (it usually converges to a slightly worse result but not significantly worse). By "270000 epochs", are you referring to epochs or steps?

mive93 commented 4 years ago

Ok thank you for your answer. Yeah, I meant 270000 step. I think the loading is correct, but I will try with changing optimizer and lr. I will let you know, Thanks!