search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.63k stars 7.95k forks source link

Custom Model detecting Objects from cocodataset #3668

Open xrv0 opened 5 years ago

xrv0 commented 5 years ago

So I trained my custom dataset and its detecting images from the cocodataset(ex: person) it was not trained on neither are they named in my .names file. When I remove the coco files(coco.data, coco.names, ...) and try to run the detect program it gives an error due to not being able to access the coco.data file.

./darknet detect cfg/mycoolproject-tiny.cfg backup/mycoolproject/mycoolproject-tiny_100.weights data/mycoolproject/100009106.jpg

(I know that 100 iterations is not enough, dont tell me)

I trained it on the pretrained darknet53.conv.74 weigth

AlexeyAB commented 5 years ago

Use detector test instead of detect

https://github.com/AlexeyAB/darknet#how-to-use-on-the-command-line

xrv0 commented 5 years ago

Thank you gonna try that. Is it normal that after 8hours of trying on a 1060 I did not get to 1000 iterations? When I try to resume the training do I just link my latest weight file as the pretrained model?