WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

AP is significantly lower than other methods,please help me

huoyiyang opened this issue · comments

Thanks for your error report and we appreciate it a lot.

Checklist

  1. I have searched related issues but cannot get the expected help.
  2. The bug has not been fixed in the latest version.

Describe the bug
I use my dataset with coco dataet format, other method such as mask rcnn , gc net ,yolact implemented by mmdetection all work well. all on colab!
but when i use in solov2 1x r50 ,there are several question!
(1)the loss_ins can not decrease , so i try to change lr, the loss can down,but still stop at around 0.3
(2)the test AP is significantly lower than other methods,about 20%
(3)i try to change the hyperparameter grid ,i enlarge them ,it seems useless

Reproduction

  1. What command or script did you run?
    !python tools/train.py configs/solov2/solov2_r50_fpn_8gpu_1x.py

  2. Did you make any modifications on the code or config? Did you understand what you have modified?
    i change the lr ,with your origin lr, the loss_ins stay=3

  3. What dataset did you use?
    my dataset ssdd ,no problem
    Environment
    i change your modified code to torch 1.4,torchcheck to atcheck

Error traceback
If applicable, paste the error trackback here.

![image](https://user-images.githubusercontent.com/73746988/122396688-d3d0d280-cfaa-11eb-89ad-f2c80b174268.png)

A placeholder for trackback.

Bug fix
If you have already identified the reason, you can provide the information here. If you are willing to create a PR to fix it, please also leave a comment here and that would be much appreciated!