Open oscar7eaglewatch opened 5 months ago
Yes, in this case we need to re-export from PyTorch without using dynamic shape so that our input and output shape will not have (-1) shape. I am so busy with my current job rn so I will try my best to release the export code asap.
Hi, Thanks for your work. I have tried to export your onnx model with TensorRT official API and it went well.
The original yolov7 exporting method to TensorRT (https://github.com/WongKinYiu/yolov7?tab=readme-ov-file#export) didn't work because your onnx model has -1 dimension as a batch size but the code (https://github.com/Linaom1214/tensorrt-python.git) requires the specific number of the batch size (1)
But when I tried to run the code (https://colab.research.google.com/github/WongKinYiu/yolov7/blob/main/tools/YOLOv7trt.ipynb#scrollTo=kRqqsjDcmyNj) of the yolov7 with export TRT file, it returns an error, it says that the (-1, 57) dimension can't be accepted.