Lingrui Gu

能通过知乎私发给我你的wechat吗？ 知乎链接：https://zhuanlan.zhihu.com/p/505992733

> > 能通过知乎私发给我你的wechat吗？ 知乎链接：https://zhuanlan.zhihu.com/p/505992733 > > I feel sorry to be here. It doesn't matter I looked at your code, too. It's great

@cydiachen I have the idea to reproduce the network on MMDet. I hope we can reproduce it together

需要几个点才能达到43.1的精度哦。 1、要8卡机器，并且要达到paddle版本中的batchsize。 2、使用imagenet的预训练权重。（目前的代码其实暂未支持加载imagenet预训练模型，需要自己手动改下代码）

Hello, thank you for you nice PR. It is nice to use built datasets instead of `pycocotools`. This allows your script to adapt to all datasets in mmyolo. Methods to...

> I did not try tensorrt on these models. However, if the layers in the models are supported by tensorrt, you can convert them. Thank you. I tried to convert...

我猜测，由于测试的患者L506的CT图是腹部图像，而腹部图像在CT中成像HU值都比较接近，一般都会使用-160~240这样的窗宽窗位，所以这里就按照这样来设置了

比较可惜，因为精力问题并未完成这部分实验。期待您，如果有相关实验，可以分享相关心得 ***@***.*** From: Jinwang Wang Date: 2024-01-16 16:58 To: open-mmlab/mmyolo CC: Lingrui Gu; Author Subject: Re: [open-mmlab/mmyolo] add clip backbone (PR #722) Can you share the accuracy comparison results between...

ppyoloe use `SiLu` in `RepVGGBlock`, not `ReLu`

`k` in `ConvModule` for `PPYOLOEBasicBlock` is 3, not 1