severe overfitting with a single GPU

[hartvon]

您好，
我在使用strong_baseline方法进行DukeMTMC-reID -> Market-1501的UDA训练时出现了严重的过拟合。

这是我对config的修改（我的设备最多支持batch size为8）：

• samples_per_gpu: 16 -> 8
• lr: 0.00035 -> 0.000035

这是我的训练log：

# performance on val dataset
Mean AP: 81.4%
CMC Scores:
  top-1          91.4%
  top-5          97.5%
  top-10         98.8%
Validating time:  0:00:09.783462
* Finished epoch  49  mAP: 81.4%  best: 81.4% *

# performance on test dataset
Mean AP: 55.4%
CMC Scores:
  top-1          78.6%
  top-5          91.6%
  top-10         94.5%

验证精度和测试精度相差很大，并且最终的测试精度与leaderboard上的差了近20%

请问我需要怎么调整以解决这一问题，非常感谢！

[mm-assistant]

Please use English or English & Chinese for issues so that we could have broader discussion.

[Emersonzc]

yxgeee/SpCL#31 (comment)