Hi, I have a yolov4-csp.weight model and I want to use deepstream with this model. I used demo_darknet2onnx.py to convert yolov4-csp to onnx and trtexec to .engine. But when I test the prediction the confidence score is dropped so much than the darknet detection and many boxes with high confidence for the same object, even though my trained yolov4-csp.weight had a quite good result with darknet native. Have you ever experienced that?
Hi, I have a yolov4-csp.weight model and I want to use deepstream with this model. I used demo_darknet2onnx.py to convert yolov4-csp to onnx and trtexec to .engine. But when I test the prediction the confidence score is dropped so much than the darknet detection and many boxes with high confidence for the same object, even though my trained yolov4-csp.weight had a quite good result with darknet native. Have you ever experienced that?