Closed Amanda-Barbara closed 3 years ago
It's on the roadmap - but I don't think it will happen anytime soon. However, there are multiple yolov5 implementations for TensorRT which are identical in implementation to the yolov4 in this repo that you can deploy exactly like its described in the README of this repo by just replacing yolov4: https://github.com/wang-xinyu/tensorrtx/tree/master/yolov5
@philipp-schmidt I will try it, thanks very much.
I have already tried yolov5 with triton https://github.com/huynhbaobk/tensorrt-triton-yolov5
@Amanda-Barbara you can try this https://github.com/tienluongngoc/yolov5_triton_inference_server
@tienluongngoc, @huynhbaobk : Can you share the config.pbtxt file that you used for the model that you used. I am using this but facing issue in postprocess function
name: "yolov5m_torch" platform: "pytorch_libtorch" max_batch_size : 0 input [ { name: "INPUT0" data_type: TYPE_FP32 dims: [1, 3, 640, 640] } ] output [ { name: "OUTPUT0" data_type: TYPE_FP32 dims: [1, 25200 , 85] } ] default_model_filename: "yolov5m.torchscript"
can you share the tutorial about https://github.com/ultralytics/yolov5 trained model with triton inference server? thanks.