周文青
> Tell me how it goes, I might be interested in merging this The code works, but the result is worse than smooth l1 (same configuration except for reg loss)....
> when i run the coco_validation.py, the error will arise~ Does anybody have this problem？ >  Have you solved that? I met the same problem.
same question, how can we improve it?
有哪个大神可以将代码移植到pytorch1.0或以上吗？
增加丰富的文档说明，例如如何自己从零开始在手机上构建图像识别项目，个人觉得opencv开启手机摄像头部分可以详细介绍一下
> 为什么我觉得你这个好慢啊 我开了十个线程等了半天才运行完 这个慢的一度让自己感觉程序是不是默默死掉了，跑了老半天才完成了几轮，没有hanlp快
**Enviroment:** > torch version: 1.13.0+cu117 torchvision version: 0.14.0+cu117 opencv version: 4.6.0 numpy version: 1.22.4 onnx version: 1.12.0 I have the same issue: ```python # original_model = models.detection.retinanet_resnet50_fpn(weights=models.detection.RetinaNet_ResNet50_FPN_Weights.DEFAULT) # do not...
同样的错误，而且POE的结果也不展示出来，但是单独进入poe的界面发现在里面已经回答问题了，可能是结果没解析出来
**OS：Win10 LATEX: TexLive 2022** I found a tricky way to solve this problem (tried install ghostscripts but failed). 1) Download and install Octave 8.1 (https://www.octave.org/) 2) Set the `gs` path...