TexasInstruments / edgeai-yolov5

YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite. Forked from https://ultralytics.com/yolov5
https://github.com/TexasInstruments/edgeai
GNU General Public License v3.0
649 stars 120 forks source link

Inconsitency in the reported results mAP@0.5 and mAP@0.5:0.95 #70

Open Michelvl92 opened 2 years ago

Michelvl92 commented 2 years ago

Hi, I use the latest version of the repo.

There is a huge inconsistency between the reported mAP@0.5 and mAP@0.5:0.95 as printed during the training versus the results.png E.g. during training, I got the following mAP@0.5 and mAP@0.5:0.95: 0.863, 0.603, while in the results.png it is reported as: 0.809, 0.863.

Furthermore, both scores also differ as reported by the xtcocotools (as included in the repo)

JoonHoonKim commented 2 years ago

https://github.com/TexasInstruments/edgeai-yolov5/blob/ae4e0e8b399959f0b60b0f555d4aafc49ae1d859/utils/plots.py#L504 Have you checked the 'plot_results' function? When learning key points, the value index number is different when the 'result.txt' file is loaded.

heylary commented 1 year ago

i have the same question,did you solve it, please

gadewegit commented 1 year ago

https://github.com/TexasInstruments/edgeai-yolov5/blob/ae4e0e8b399959f0b60b0f555d4aafc49ae1d859/utils/plots.py#L504

你检查过'plot_results'函数了吗?学习关键点时,加载'result.txt'文件时,值索引号不同。

I also met this problem, how to solve it,I hope to get your help.