DOTA_devkit icon indicating copy to clipboard operation
DOTA_devkit copied to clipboard
verified publisher icon CAPTAIN-WHU

Usage of `poly_nms_gpu`

Open DrRyanHuang opened this issue 3 years ago • 4 comments

Hi, thx for your very helpful code. Could you please give a demo of poly_nms_gpu?

DrRyanHuang avatar Aug 25 '22 02:08 DrRyanHuang

Hello, can you explain how to use it please?

yoloyash avatar May 10 '23 18:05 yoloyash

The function can be found at this link https://github.com/CAPTAIN-WHU/DOTA_devkit/blob/master/poly_nms_gpu/nms_wrapper.py The interface of poly_nms_gpu is the same as py_cpu_nms_poly https://github.com/CAPTAIN-WHU/DOTA_devkit/blob/99388551054be9a6dabb01c8bb2a7eb562d57b4f/ResultMerge.py#LL17C1-L17C1

dingjiansw101 avatar May 11 '23 03:05 dingjiansw101

@dingjiansw101 Thanks a lot for your reply. It's working now. However, it's very slow for me. Is there anything I'm doing wrong or missing out? Trying to follow yolov5_obb implementation and the nms per batch exceeds 50-60 seconds in some cases. I'm using it as follows. [cpu:Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz, gpu: Nvidia Tesla T4] from poly_nms_gpu import poly_nms poly_nms.poly_gpu_nms(dets, thresh=0.5, device_id=0)

yoloyash avatar May 11 '23 06:05 yoloyash

Actually, I recommend you use py_cpu_nms_poly. GPU version is not always faster.

dingjiansw101 avatar May 11 '23 06:05 dingjiansw101