Bendik Samseth comments

Results 11 comments of


                                            Bendik Samseth

设置m_config.mode = 2，你可以看看我的案例，而且需要配置校准表

look https://github.com/ttanzhiqiang/onnx_tensorrt_project/blob/main/model/darknet_onnx_tensorrt_yolo/yolov3.cfg

Using netron, view the network. In your yolo Plugin，add down_stride(down_stride = (input / output_feature)).

sorry,this project must be on Windows.I'll do it on ubuntu in the future.

I build onnx-tensorrt，because I use centernet.if you no use centernet,you unnecessary build onnx-tensorrt.

TensorRT 7.2.2.3

你需要重新生成onnx（batch=2），然后再使用这个代码生成trt文件

看看你的python代码的输入和输出，与Tensorrt的输入输出是否一样

yolox模型的anchor机制使用了自动求anchor的算法，和之前的yolov5是不一样的，你可以参考一下我写的代码

this project include Yolov5(4.0)/Yolov5(5.0)/YoloR/YoloX/Yolov4/Yolov3/CenterNet/CenterFace/RetinaFace/Classify/Unet