AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.65k stars 7.96k forks source link

mAP calculation for final weights #3088

Open ghost opened 5 years ago

ghost commented 5 years ago

Hi,

I tried to calculate the mAP for final weight e.g. yolov3-mio_final weights using the following command, but all mAPs are zero. But it is ok for other weights like yolov3-mio_10000.weights.

./darknet detector map cfg/mio.data cfg/yolov3-mio.cfg backup/yolov3-mio_final.weights

I want to test other weights e.g. yolov3.weights or yolov3-spp.weights on my validation set and I have the same problem. Zero mAP for all classes.

Thanks,

AlexeyAB commented 5 years ago

@behnamj Hi,

I tried to calculate the mAP for final weight e.g. yolov3-mio_final weights using the following command, but all mAPs are zero. But it is ok for other weights like yolov3-mio_10000.weights.

Check that you used identical commands. I can't reproduce this issue.

I want to test other weights e.g. yolov3.weights or yolov3-spp.weights on my validation set and I have the same problem. Zero mAP for all classes.

What command do you use and can you show screenshot of result?

ghost commented 5 years ago

Hi again, The first command is: ./darknet detector map cfg/mio-fine.data cfg/yolov3-mio-fine.cfg backup/mio-fine/yolov3-mio-fine_final.weights The output is: Screenshot from 2019-05-06 10-52-20

The second command is: ./darknet detector map cfg/mio-fine.data cfg/yolov3-mio-fine.cfg yolov3-spp.weights The output is: Screenshot from 2019-05-06 10-53-25

The third command is: ./darknet detector map cfg/mio-fine.data cfg/yolov3-mio-fine.cfg backup/mio-fine/yolov3-mio-fine_15000.weights The output is: Screenshot from 2019-05-06 10-54-42

AlexeyAB commented 5 years ago

./darknet detector map cfg/mio-fine.data cfg/yolov3-mio-fine.cfg backup/mio-fine/yolov3-mio-fine_final.weights

It seems that backup/mio-fine/yolov3-mio-fine_final.weights was trained for different yolov3-mio-fine.cfg model.