Hao Yuan

bifpn中加权聚合时（简单的注意力机制），权重参数不更新问题

10

作者您好，我尝试将您这个代码中bifpn中的加权聚合方法移植到别的目标检测框架中看看能不能有作用，我修改后代码成功跑起来了，但在查看训练好的模型字典时，发现权重w初始化值是【1,1】训练好的模型，w仍然是【1,1】，希望大神能帮我看下我的代码是不是哪出现了问题，从而导致该参数完全不更新。  

prt file problem

4

Under your guidance, I successfully ran the code without errors. But I encountered some problems when I ran prt_util.py to generate prt file. Here's problem: when I ran the 'prt_util.py'...

I uesd the **Partial_conv** in my model, which includes the **Channel Split(torch.split)**. In the inference(after **ModelSpeedup**), i got the error in `x1, x2 = torch.split(x, [self.dim_conv3, self.dim_untouched], dim=1)`: RuntimeError: start...

other public datasets，overfitting issue

6

Hello, I am trying to adapt the MPII_3DHP dataset on your code, but it's totally overfitting First I train it on H36M+MPII，the MPJPE is about 50+. This is normal Then...

Keypoints number issues

1

Hello, I have currently created an SMPL model using demo. Besides I would like to use this SMPL model to obtain the coordinates of the 3Dkeypoints of the human body,...

issue about preprocess amass dataset

6

Hi, thank you for providing such an exciting project ! I met some problems in running preprocess_amass.py. It need smplh and dmpls model. I got smplh model data from [MANO](https://mano.is.tue.mpg.de/index.html)，...

### What is the feature? 我参照mmpose以及simcc的代码在自己的三维姿态估计项目中搭建了分类头，xyz可取值都是0-255(256类)，但训练出来的结果却是不对称的，x正常收敛，y和z容易出现完全不收敛的情况。这种多分类头任务收敛不均匀是常见的现象么？应该如何如何解决？  ### Any other context? _No response_

I tried to add variance prediction in my own loss, so I tried to reproduce step by step. so, at the first step, i just added the var prediction in...