TensorRT-CenterNet
TensorRT-CenterNet copied to clipboard
Different result form pytorch and tensorrt
Thanks for your great work. I have test the model for pytorch and tensorrt which trained on my dataset. The dataset only have 7 classes.
This is result form pytorch model.
and this the tesorrt model conterted from onnx.
It seems that tensorrt mode has more small object on the top of image, and also have some overlap bbox. Is there any way to solve this problem? thanks!
The small objects disappeared when I changed the classNum and className. But the result still contains overlap.
I think this is due to no nms implementation as specified in Readme file