search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.72k stars 7.96k forks source link

mAP function shows andefine object ( yolov3 train)? #3238

Open Modeinpy opened 5 years ago

Modeinpy commented 5 years ago

Hi. I am trying to train custom object detection model..When I am launching training with -map attribute, it fails in 1k iterations with segmentation fault error. I have noticed, that map result shows more objects when I defined in obj.names file (file has 4 names), but map results shows 5. I can't find is it labelling problem (I have double triple checked labelling (labelled with yolo_mark). If I launch training without map function, it goes forward good, any errors are showed. P.S. some validation pictures doesn't have labelling file, is that might be problem?

AlexeyAB commented 5 years ago

@Modeinpy Hi,

I have noticed, that map result shows more objects when I defined in obj.names file (file has 4 names), but map results shows 5.

Show screenshots of

  1. results shows 5.
  2. obj.data file
  3. obj.names file
  4. command that you uses for checking map
Modeinpy commented 5 years ago

61833376_418282859014788_6097498253324976128_n 61856364_2690457430968880_4150859546807500800_n 61982154_619284605257263_3771356446986338304_n 62242456_418146349030084_3074400211099975680_n

http://www.ee.oulu.fi/~olli/Projects/Lumber.Grading.html?fbclid=IwAR3esjyQXrH701P1Cf21jo4dh9wcdxjN7AzydKxVJ-vqQifMoeP9uejGVdI

AlexeyAB commented 5 years ago

You set classes=5 in last [yolo] layer in your cfg-file.