The visualization of evaluation results is messy

[donkehuang]

Hi, Thanks for your great work and effort for releasing the code. When i tried to reproduce the results as the paper said, i had the following trouble and hard to handle.

The evaluation results cannot be reproduced when i used the pretrained resnet weight and the results in visualization is messy. Here is the command and results i tried.

• Evaluation
  CUDA_VISIBLE_DEVICES=0 python tools/test.py projects/configs/VAD/VAD_base_stage_2.py ckpts/resnet50-19c8e357.pth --launcher none --eval bbox --tmpdir tmp
• Visualization
  

The img_norm_cfg version mentioned in issue#9 is also tried, but the results are still the same.

[PhilWallace]

I think the parameter after the config file should be the VAD model instead of the resnet50 model.

python tools/test.py projects/configs/VAD/VAD_tiny_stage_2.py ./ckpts/VAD_tiny.pth --launcher none --eval bbox --tmpdir tmp