CodingMice

> > @ZHANG-SHI-CHANG 谢谢你的分享 > > 人脸检测你也可以直接用dlib，但是关键点最好用FAN，我测试dlib，FAN和自己的关键点模型，FAN表现的最好，但也没有3ddfa稳定，我想这个输入语义的影响还是蛮大的 我测试了3ddfa和2dasl，但感觉对于视频输入都不太稳定，有什么好的解决思路吗？

> > > > @ZHANG-SHI-CHANG 谢谢你的分享 > > > > > > > > > 人脸检测你也可以直接用dlib，但是关键点最好用FAN，我测试dlib，FAN和自己的关键点模型，FAN表现的最好，但也没有3ddfa稳定，我想这个输入语义的影响还是蛮大的 > > > > > > 我测试了3ddfa和2dasl，但感觉对于视频输入都不太稳定，有什么好的解决思路吗？ > > 如果是抖动问题就使用平滑策略吧，你的测试效果哪个好一些呢 是平滑检测框吗？我测试的时候大侧脸抖动的比较明显。单凭肉眼看也看不太出来哪个更好些，有什么指标可以衡量3Dmesh抖动程度的么？

> > > @et148 No，first，the dataloader should read the txt，I tried to guess and pass. But the author use the vgg net in net.py, and it must read the conv...

> > > @CodingMice > > I downloaded and tried，but I think if need to match the author‘s network it need more work. >  > ’ I have run...

> > > @CodingMice > Thank you！By the way，can you use multi-gpu training? I have tried many times but still failed I do not use the muti-gpu training. If I...

> @CodingMice Thank you so muc. I can run it successfully. I am not very familiar with pytorch. I tried to train the model with my own database. But it...

> > > That may be a little difference with the statement in paper. You can also try to implement yourself and make a PR. Hi, I have implemented makeuploss...

> > > > > That may be a little difference with the statement in paper. You can also try to implement yourself and make a PR. > > >...

> > > > > > > That may be a little difference with the statement in paper. You can also try to implement yourself and make a PR. >...

> @CodingMice 可以试一下mnn的inference，那个onnx我应该是放错模型了，回头我更新一下。 好的，坐等提供模型进行测试