realtime-semantic-segmentation-pytorch
realtime-semantic-segmentation-pytorch copied to clipboard
发现一处数值异常的地方,还请检查一下。
https://github.com/zh320/realtime-semantic-segmentation-pytorch?tab=readme-ov-file#full-resolution-on-cityscapes
这张表中,看到了DDRNET的FPS数有明显异常 233FPS,别的模型参数量比它小很多的也没有它快。
我找了官方的测试结果图
我们以BiSeNetv2作为参考,图中大概在150FPS,DDRnet的FPS在110左右。
而在你的表格中 BiSeNetv2 的FPS比较接近,而ddrnet的fps为233。我心生疑虑,还请勘误。
Hi, thank you for pointing out the abnormal FPS for DDRNet. I have also noticed this and double checked on my machine. Currently I can not provide you a solid explanation about it. Maybe there is some issue about my machine? I am not sure.
Do you have available graphics card and would you please run test_speed.py and report your performances here?
Best,