PyTorch_YOLOv4 icon indicating copy to clipboard operation
PyTorch_YOLOv4 copied to clipboard

测试集精度几乎为零

Open haoxue1215 opened this issue 3 years ago • 2 comments

image Scanning labels D:\Anaconda3\envs\pytorch\ayolo\pytorch_YOLOv4\VOCdevkit\VOC2019\JPEGImages.cache (864 found, 0 missing, 0 empty, 0 duplicate, for 864 images): 100%|█| 864/864 Class Images Targets P R [email protected] [email protected]:.95: 100%|███████████████████████████████████████████| 216/216 [02:43<00:00, 1.32it/s] all 864 3.84e+04 7.08e-05 0.000124 5.35e-07 2.09e-07 pedestrian 864 1.09e+04 0 0 0 0 car 864 2.22e+04 0.000283 0.000497 1.55e-06 7.35e-07 van 864 4.72e+03 0 0 5.86e-07 9.94e-08 tricycle 864 621 0 0 0 0 为什么自己训练好的模型用来测试的时候精度几乎为零啊,有什么需要注意的地方吗

haoxue1215 avatar Mar 02 '21 07:03 haoxue1215

信息量不足以分析原因.

目前有發現 mish_cuda 在某些cuda/pytorch/gpu上運行會有問題.

WongKinYiu avatar Mar 04 '21 01:03 WongKinYiu

我也遇到同样问题,请问解决了吗

forwardwfg avatar Apr 20 '22 06:04 forwardwfg