It is a simple tool based on mmdetection2.0. The tool is used for visualize test result produced by mmdetection 2.0, including drawing PR curves and loss curves.(PR曲线绘制,loss曲线绘制)
The program supports drawing two figures. The loss figure includes six kinds of training loss. The PR figure includes PR curve and F1-score.
loss figure:
loss_rpn_bbox
loss_rpn_cls
loss_bbox
loss_cls
loss
acc
PR figure
PR_curve
F-measure
git clone https://github.com/xiaoyu1233/mmdetection2.0_visualize
put visualize.py
under /mmdetection/tools/
put voc_eval_visualize.py
under /mmdetection/tools/
put mean_ap_visualize.py
under mmdetection/mmdet/core/evaluation/
When training finished, you will have work _dirs directory in your mmdetection directory. Find your latest json directory and copy the path. And use it in this command.
At the same time, you should prepare a directory in your mmdetection diretory. Create output/work_dirs/faster_rcnn_r50_fpn_1x_voc directory.(The diretory in your output folder should be same as your work_dirs, or you can change it in code by yourself).
python tools/visualize.py work_dirs/faster_rcnn/faster_rcnn_r50_fpn_1x_voc/xxxxxxxx.log.json(your latest path)
Then check the output/work_dirs/faster_rcnn_r50_fpn_1x_voc directory, you will get the loss figure.
Make sure voc_eval_visualize.py and mean_ap_visualize.py settled down And prepare a directory in your mmdetection diretory. Create mmdetection in your mmdetection directory. Use this command
python tools/voc_eval_visualize.py ./result.pkl(your pkl file path) ./configs/faster_rcnn/faster_rcnn_r50_fpn_1x_voc.py(your config file)
Then check mmdetection/mmdetection directory, you will get PR figure.
if issues are not solved in time, welcome to talk with me at my CSDN blog: https://blog.csdn.net/AI414010