YOLOv3 custom data training model having an error when training pre-trained weights on custom data Using Darknet framework with the darknet53.conv.74 pre-trained model and yolov3.cfg that i edited to suit my custom data:
example of what my .txt file for the bounding box looks like: (0 0.336232 0.320863 0.052174 0.515108), i have only 1 class so it's 0.
I have changed the classes number to 1, filters to 18((classes+5)*3), batches 4000(minimum), min/max batches(3800-4200) and of course changed all the 3 [Yolo] layers with their convolutional layers, Height = Width, etc..
I'm working on google.collab but i edit my files on my machine which is windows.
I checked for the CR - LF, using windows and my files contain "\n" so i did not change them assuming it works fine with a windows machine.
i have my test and train files filled, labelled_data.data, "!file" compiled correctly, checked the path !darknet/darknet detector train custom_data/labelled_data.data darknet/cfg/yolov3_custom.cfg custom_weight/darknet53.conv.74 -dont_show
assuming i took care of all the details and everything will work fine.
BUT,i keep getting this error when i try t o train the model:
yolov3_custom
First section must be [net] or [network]: No such file or directory
darknet: ./src/utils.c:256: error: Assertion `0' failed.
I don't know what could've gone wrong with this!
if you create your files in windows and try to run in linux machine, you should change your encode using dos2unix library
and then :
this worked for me