mirrors
/
mmyolo
mirror of https://github.com/open-mmlab/mmyolo.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
mmyolo/projects/assigner_visualization
History
yechenzhi 75618020f8
RTMDet Assigner visualization (#528) 
* fix format

* return multiple pos assigns

* rewrite to get matched_gt_inds

* ignore corrupted images

* rm RTMDetectorAssigner

* fix bug for different devices

* add warnings when use rtmdet without checkpoint

* add priors for rtmdet

* fix format

* add readme

* fix format

* fix readme and typo

* typo

* fix note
 		2023-02-13 11:42:11 +08:00
..
configs RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00
dense_heads RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00
detectors RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00
visualization RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00
README.md RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00
assigner_visualization.py RTMDet Assigner visualization (#528) 2023-02-13 11:42:11 +08:00

README.md

MMYOLO Model Assigner Visualization

Introduction

This project is developed for easily showing assigning results. The script allows users to analyze where and how many positive samples each gt is assigned in the image.

Now, the script supports YOLOv5 and RTMDet.

Usage

Command

python projects/assigner_visualization/assigner_visualization.py projects/assigner_visualization/configs/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_assignervisualization.py

Note: YOLOv5 does not need to load the trained weights.

python projects/assigner_visualization/assigner_visualization.py projects/assigner_visualization/configs/rtmdet_s_syncbn_fast_8xb32-300e_coco_assignervisualization.py -c ${checkpont}

${checkpont} is the checkpont file path. Dynamic label assignment is used in RTMDet, model weights will affect the positive sample allocation results, so it is recommended to load the trained model weights.

If you want to know details about label assignment, you can check the documentation.