HaronCHou

 This thresh is to avoid obvious mismatches, first filter out mismatches with a large distance. In a row, find(row > thresh). In matlab offcical code, this value set to...

I'm also encountering this error. @dnth 's solution worked for me.

作者有docker提供的

confidence_threshold和nms_threshold是成员变量，给开放一个接口修改，就可以了。

https://github.com/WZMIAOMIAO/deep-learning-for-image-processing/issues/740#issuecomment-1649253697 参考这个呢，注意pascal_voc_classes.json文件夹里面的内容；RetinaNet和前面的网络结构不一样，类别是从0开始，不+1的。

yolov3-spp中有giou diou ciou的计算代码

lr不是固定的吧？是有函数的呢，有学习率update。up用的Cos函数的lr下降呀。`scheduler.step()`

> 1850行处：const int MAX_IMAGE_BBOX = 1024 1024设的太小了，会出现漏检的情况，应该开放出来或设大点 为啥呢

> 你好，请问使用fp16需要修改哪些地方吗。还是说和fp32一样的？ type还是mode给FP16就可以了

``` (base) root@f3cd9dc49bd1:/datav/projects/tensorRT_Pro/simple_yolo/workspace# /datav/projects/tensorRT_Pro/simple_yolo/workspace/pro ===================== test YoloV5 FP32 yolov5s2 ================================== [info][simple_yolo.cu:2216]:Compile FP32 yolov5s2_dynamic.onnx. [warn][simple_yolo.cu:1116]:NVInfer: onnx2trt_utils.cpp:364: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support...