regainOWO

Serving部署的模型是yolov3_darknet53_270e_coco, paddle_serving_client和paddle_serving_app版本为0.9.0 **在windows下使用http方式获取服务**，参考[resnet50_http_client.py](https://github.com/PaddlePaddle/Serving/blob/v0.9.0/examples/C++/PaddleClas/imagenet/resnet50_http_client.py)和[test_client.py](https://github.com/PaddlePaddle/Serving/blob/v0.9.0/examples/C++/PaddleDetection/yolov3_darknet53_270e_coco/test_client.py) 转化方法比较暴力，但可用 ``` python import sys import numpy as np from paddle_serving_client import HttpClient from paddle_serving_app.reader import * import cv2 def main(argv): preprocess = Sequential([ File2Image(), BGR2RGB(), Resize(...

@ppogg PicoDet-s.yaml 相比PicoDet-m.yaml PicoDet-l.yaml PicoDet-x.yaml有些区别 https://github.com/ppogg/YOLOv5-Lite/blob/ca7ed7ca0bb578fe6e5eaa777e84f661ad457e49/models/hub/PicoDet-s.yaml?_pjax=%23js-repo-pjax-container%2C%20div%5Bitemtype%3D%22http%3A%2F%2Fschema.org%2FSoftwareSourceCode%22%5D%20main%2C%20%5Bdata-pjax-container%5D#L45 s用的是Add，其他的则是用的concat，有什么区别吗？

@hukaixuan19970627 我觉得可以多添加一个修改后的文件来解决这个问题，针对不同的pytorch版本编译不同的文件，我的对setup.py做了一些修改，函数`make_cuda_ext`多添加了一个参数`sources_cuda_later`会根据不同的pytorch版本，编译不同的文件。 ``` python #!/usr/bin/env python import os from setuptools import setup import torch from torch.utils.cpp_extension import (BuildExtension, CppExtension, CUDAExtension) def make_cuda_ext(name, module, sources, sources_cuda=[], sources_cuda_later=[]): define_macros = [] extra_compile_args...

@safaATcurrus Thank you for your advice. i found the memory cost is much small than before, but the cpu cost in my laptop likely abou 9%, the example is here:...

> Output shape from yolov5 should be (batch size, number of bboxes, 4 bbox values + 1 objectness + confidence for each class) but in yolov8 it is (batch size,...

@JustasBart highly recommend [hukaixuan19970627/yolov5_obb](https://github.com/hukaixuan19970627/yolov5_obb) . it use increase 180 dim for no in Detect to classify bounding box angle, not a regression method, referece the [paper](https://arxiv.org/abs/2003.05597v2). also the repo [open-mmlab/mmrotate](https://github.com/open-mmlab/mmrotate)...

i think the [issues 151](https://github.com/hukaixuan19970627/yolov5_obb/issues/151#issuecomment-1515600373) may help to fix it

> 感谢您的恢复，我用低版本的mmcv\mmdet进行了代码运行，但是我又遇到了以下问题： Traceback (most recent call last): File "tools/train.py", line 153, in main() File "tools/train.py", line 142, in main train_detector( File "/data4/libing/bisai/OBBDetection/mmdet/apis/train.py", line 133, in train_detector runner.run(data_loaders, cfg.workflow, cfg.total_epochs) File...

在你的训练相关的配置文件exps/example/yolox_obb/yolox_s_dota1_0.py下，类Exp的成员变量self.no_eval = False，epoch设大一点，默认好像是10个epoch然后验证一次，具体可以看类Exp的父类OBBExp的成员变量self.eval_interval的值。

@D-EEPlearning 字面意思 no eval 不进行验证，设为False就是指进行验证。这个参数貌似是作者自己加的，原yolox的仓库里没有。