paleomoon

One can use onnx2ncnn or pnnx to convert yolov5 pytorch model **without Detect() layer** to ncnn, that is export with **--train** option. If use pnnx, you have no need to...

+1，第三方socket工具作为服务端能收到数据，将收到的数据（字节数组）再发给客户端，客户端收不到，onSocketReadResponse不执行。协议没修改，使用的是默认的（4字节header+body）。 比如发送{“cmd”:0}，对应字节数组[00 00 00 09 7B 22 63 6D 64 22 3A 30 7D]，socket工具收到后，回复同样的字节数组给客户端（所以协议应该是没问题的），客户端收不到。但使用demo作为服务端能正常收发。 这个问题挺多人遇到了，为什么这么久了还没有答案？

Just add `grad_clip` in train.yaml as vivek87799 said, now loss is decreasing. But I don't understand how this works.

This happened when I made a mistake in CMakeList.txt: `set(CMAKE_BUILD_TYPE DEBUG)`. Change to `set(CMAKE_BUILD_TYPE Debug)` fixed the issue 😂.

> https://github.com/tensorlayer/hyperpose/issues/323#issuecomment-704676488 Same problem with lopps-resnet50-V2-HW=368x432.onnx: Input shape mismatch, other models is ok.

> @paleomoon > > The lopps being 368x432 means actually 432x368. Flip your dimensions and should be ok. Sadly it's not ok. The problem is not here.

As a beginner, I think pad in yolov3.cfg means whether to use padding. In yolo, only stride controls the output size. If stride = 1, output the same size as...

请问你解决了吗？ckpt.t7明明是512维输出，代码里咋设成256了。。。

@abhigoku10 Hi, first you can convert to onnx: https://openpifpaf.github.io/dev/cli_help.html#export-onnx, then use trtexec command tool to convert onnx to tensorrt. Like this: ``` python -m openpifpaf.export_onnx --input-width [INPUT_WIDTH] --input-height [INPUT_HEIGHT] --checkpoint...

Hi, any update? the C++ implementation of this repo is bytetrack.