EthanW-coder
commented
1 week ago 您好，感谢您将 Mamba 带到 yolo 系列方面的出色工作，但是当我在 widerface + 私人数据上训练 Mamba-yolo-T 时，结果低于 yolov8-n。
曼巴-约洛-T
约洛夫8-N
我只是换成，你有什么好的建议吗？
Detect head``Pose head
Thank you for your great interest in our work, we trained on COCO and VOC and reported it in the paper, in fact we also tried training on some downstream tasks such as CrowdHuman and CityScapes all with some relatively good results, so far nothing like what happened during your experiments. In fact, you could also try using pure Mamba in combination with YOLOv8 to explore better results in downstream tasks by further ablation experiments, but of course this is just our suggestion. However, your experiment is very valuable, which will help us to further optimize the whole Mamba architecture in the exploration of detection. In the future, we will introduce some more suitable Mamba-YOLO for downstream tasks and update it in the code base, please pay attention to it.
xsa12345
Yshelgi
try to update gcc version
Hello and thank you for your good job in bringing Mamba to the yolo-series,but when i train Mamba-yolo-T on widerface + private data,the results are lower than yolov8-n.
Mamba-yolo-T
yolov8-n
I just change
Detect headtoPose head,do you have any good advice?