search

WongKinYiu / yolor

implementation of paper - You Only Learn One Representation: Unified Network for Multiple Tasks (https://arxiv.org/abs/2105.04206)
GNU General Public License v3.0
1.98k stars 524 forks source link

Incorrect precision and recall #74

Open madr3z opened 2 years ago

madr3z commented 2 years ago

Precision is either 1 or 0 and recall is always 0.000589 in all epochs even after 300 epochs. I am using a single class dataset (binary detection) and I am guessing that this issue happens only with single class datasets. I tried copying yolov5 repo's utils/metrics.py but still getting the same problem. This issue also happened in Scaled-Yolov4 repo.

yusiyoh commented 2 years ago

I have also strange test.py results, which shows very low recall (check off class) but very high mAP, I don't know how. image

TheConstant3 commented 2 years ago

i'm faced with same problem while training! Снимок экрана от 2021-09-01 02-11-14

yusiyoh commented 2 years ago

i'm faced with same problem while training! Снимок экрана от 2021-09-01 02-11-14

You may try to clone repo again since the metrics.py is changed. However, I am not sure about precision and recall. mAP will definitely change.

dhivakark commented 2 years ago

Facing the same issue image

qinghezeng commented 2 years ago

Same here! And the step is (13 * actual epoch + 6) ... image

nhatanh174 commented 2 years ago

Hi! I'm faced same issue. Anyone help me solve this problem ? Thanks.