search

voldemortX / DST-CBC

Implementation of our Pattern Recognition paper "DMT: Dynamic Mutual Training for Semi-Supervised Learning"
BSD 3-Clause "New" or "Revised" License
135 stars 17 forks source link

About using a better model #11

Closed wing212 closed 2 years ago

wing212 commented 2 years ago

I would like to ask if you have used a better model for experimentation, such as deeplab V3+. Will it bring better accuracy if you use a better model?

voldemortX commented 2 years ago

I would like to ask if you have used a better model for experimentation, such as deeplab V3+. Will it bring better accuracy if you use a better model?

I would say it is highly likely we can get better performance with a better model.

However, when this work finished first draft (that's like early 2020), the SSS community commonly uses DeepLabV2 (if not only DeepLabV2). I never tried other models myself (I believe in SSL, absolute performance mean little, compared to gap to Oracle or Baseline), you are encouraged to try it though, I have some better segmentation models implemented in a similar style at https://github.com/voldemortX/pytorch-auto-drive.

The same issue was raised in #1

voldemortX commented 2 years ago

@wing212 Closing as a duplicate of #1.

wing212 commented 2 years ago

Thank you for your reply, I will take a closer look