How to extract classes, confidences and boxes from yolov4 onnx output

linghu8812 / tensorrt_inference

708 stars 208 forks source link

How to extract classes, confidences and boxes from yolov4 onnx output #92

Open marcusturewicz opened 3 years ago

marcusturewicz commented 3 years ago

I have successfully converted a Darknet Yolov4 model to onnx using export_onnx.py.

However, I don't understand the output format. I have 4 classes and my model's output is 1 x 68607 x 9. How can I extract the classes, confidences and boxes from this output?

Ram-Godavarthi commented 3 years ago

Were you able to get the expected outputs with your model ?? For, me batch 64 works (very slow). but batch size = 1 -> does not not work at all. Getting wrong bounding boxes. Should we change any calculation part in the yolov4.cpp file ??