training process killed

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.58k stars 7.95k forks source link

training process killed #2532

Open dengsicong opened 5 years ago

dengsicong commented 5 years ago

After runing in cmd: \darknet-master\build\darknet\x64>darknet_no_gpu.exe detector train data/obj.data yolo-obj.cfg darknet53.conv.74 the training progress is killed without any errors. default

AlexeyAB commented 5 years ago

Are there bad.list or bad_label.list files?
Can you successfully train with random=0?
How many CPU RAM do you have?

dengsicong commented 5 years ago

Thanks for you reply. Yes,there is a bad.list.But the sequence can not match my train.txt. And the same result happened by training with random=0. My CPU RAM 8GB(7.85GB usable)

default my document: \darknet-master -Makefile \build\darknet\X64: -data --obj ---img1.png ---img1.txt ...... ---img24.png ---img24.txt --obj.data --obj.names --train.txt

-bad.list -darknet_no_gpu.exe -yolo-obj.cfg

AlexeyAB commented 5 years ago

Remove bad.list & bad_label.list, train again, and then look at bad.list & bad_label.list again.

dengsicong commented 5 years ago

Thanks.This problem has been solved.Change opencv debug to release can start training .Bus there is a new problem,the image of average loss can not respond. And Used FMA&AVX2 appeard in cmd default default

dengsicong commented 5 years ago

And if i remove bad.list,after starting training,there is no bad.list existing.And from starting,there is not bad_label.list,

gongjing1996 commented 5 years ago

Thanks.This problem has been solved.Change opencv debug to release can start training .Bus there is a new problem,the image of average loss can not respond. And Used FMA&AVX2 appeard in cmd

What means changing opencv debug to release? The training is conducted with command window,could you tell me how to change in detail? Thank you very much!