2 quick Questions for SOT inference: IOU metric? Show ground truth bbox in .mp4 output?

[santideleon]

I am using sot pretrained models of stark and siamese. I am using jpg images numbered 1 to n and a .txt file for my ground truth bbox with each row representing a frame and in the format (max_x, max_y, min_x, min_y). my code is along the lines of this: !python demo/demo_sot.py
/content/mmdetection/mmtracking/configs/sot/siamese_rpn/siamese_rpn_r50_20e_otb100.py
--input /content/biker/img
--checkpoint checkpoints/siamese_rpn_r50_20e_lasot_20220420_181845-dd0f151e.pth
--output biker_siamese_otb.mp4
--device cuda:0
--gt_bbox_file /content/biker/groundtruth_rect.txt --fps 30

I was wondering if there is already an implementation or easy way to do the 2 following things:

1. Find the average IOU of the estimated bboxes of the model.
2. In the same output .mp4 video show the ground truth bboxes as well

[dyhBUPT]

Hi, there is no codes for the average IOU evaluation. You need to implement it by yourself.

For the visualization with gt, you are recommended to refer to the demo_sot.py and modify the BaseSingleObjectTracker.show_result to support the visualization of gt.