maoxuli / darknet_det Goto Github PK

The code is tested and working well with pre-trained weights on yolov3 416 or 608 networks, with Jetson TX2 on-board CSI camera. With power mode 1 (lowest power mode), 416 network output is about 3 fps, while 608 network output is only about 1.3 fps. I downloaded yolov3-tiny network and tried, the output is about 13 fps. But the output bounding boxes are not correct, way some smaller than the objects. It seems that the parsing code for network output has issues, or the tiny network has different detection output.