search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.76k stars 7.96k forks source link

Prediction Issue #1612

Open dexception opened 6 years ago

dexception commented 6 years ago

@AlexeyAB

I have created a dataset to detect number plates. After tagging around 12,000 images and 57,237 iterations the prediction is off target. All the prediction is above the target.

 57237: 0.212793, 0.181010 avg loss, 0.001000 rate, 0.459194 seconds, 3663168 images
Loaded: 0.248076 seconds
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000001, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.877086, Class: 0.999895, Obj: 0.826524, No Obj: 0.000556, .5R: 1.000000, .75R: 1.000000,  count: 9
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000004, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.767350, Class: 0.999836, Obj: 0.887833, No Obj: 0.000651, .5R: 1.000000, .75R: 0.750000,  count: 12
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000000, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.822426, Class: 0.999928, Obj: 0.890275, No Obj: 0.000579, .5R: 1.000000, .75R: 0.833333,  count: 12
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000001, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.790765, Class: 0.999721, Obj: 0.819195, No Obj: 0.000476, .5R: 1.000000, .75R: 0.700000,  count: 10
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000001, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.751157, Class: 0.999635, Obj: 0.825129, No Obj: 0.000675, .5R: 1.000000, .75R: 0.666667,  count: 12
Region 16 Avg IOU: 0.836096, Class: 0.997474, Obj: 0.379923, No Obj: 0.000081, .5R: 1.000000, .75R: 1.000000,  count: 1
Region 23 Avg IOU: 0.845406, Class: 0.999731, Obj: 0.921092, No Obj: 0.000657, .5R: 1.000000, .75R: 0.928571,  count: 14
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000000, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.825143, Class: 0.999936, Obj: 0.916565, No Obj: 0.000696, .5R: 1.000000, .75R: 0.769231,  count: 13
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.000001, .5R: -nan, .75R: -nan,  count: 0
Region 23 Avg IOU: 0.790452, Class: 0.999876, Obj: 0.833961, No Obj: 0.000715, .5R: 1.000000, .75R: 0.833333,  count: 12

prediction1 prediction2 prediction3

dexception commented 6 years ago

@AlexeyAB Solved !

The c++ wrapper is giving correct output. std::vector<bbox_t> box = detector.detect(clone);

This is worth investigating.