This repo has been updated to recent version.More details and annotations can be found in code.
Annotation and Specification for yolov3
Attention:Inputs should be labeld the same as COCO (yolo format):normalized && cxywh
refer to:https://github.com/ultralytics/yolov3
download yolov3.weights trained on COCO:https://pjreddie.com/darknet/yolo/
run python detect.py
You can test it on offerred image in /data/dataset
run python train.py
test on the same image to verify it feasibility.
This is my loss curls with testing on single picture for 100 epochs,it does converge well.
Run train-vis.py
to monitor loss.