Inference

[setarekhosravi]

Very bad results, when running video_demo.py.

[goutamyg]

Thank you for the feedback. Can you share more information on the video/s that you have tested? (e.g., how many videos, any challenging factor/attribute in those videos, etc). Meanwhile, I will double-check if there is any bug in the code.

[setarekhosravi]

I have tested MVT on about 6 InfraRed videos for small object-tracking tasks in InfraRed imagery. Unlike SMAT, MVT tracks objects in 5 frames, hardly.

[goutamyg]

The tracker is trained on RGB images, hence it may not work well on infrared image-based applications. That could the reason for poor performance of MVT

[setarekhosravi]

@goutamyg I know but I had tested SMAT on IR videos and its performance was much better(about 20%-30%) than MVT.