search

scutpaul / DANet

30 stars 3 forks source link

Question on the online learning for testing #1

Closed Reagan1311 closed 3 years ago

Reagan1311 commented 3 years ago

Hi, thanks for sharing the code. I have tested the provided pre-trained model in group 0, and got results with F: 40.22 & J: 41.54, which are close to the results in Table 1.

I guess these are the results without online learning. However, when I implement online learning with the default setting to finetune the result, the performance degrades a lot (F: 26.18 & J: 8.14). Why would this happen?

I use the following command: python test_DAN.py --test_best --finetune --group 1

scutpaul commented 3 years ago

In online learning, we only finetune and test for one class at a time, so we need to specify the class idx. The setting parameter is --finetune_idx. The result of your test is the result of default class idx 1 in group 1. In addition, you can set different finetune parameters for different classes to get better results.