Hao Yuan

好的那我详细说一下，我训练了100多个epoch之后然后将保存好的模型使用torch.load()函数导出，来查看代码图里的所示的w1，w2，w3，w4是否有变化，具体代码如下图  打印结果如下图 

是的是按照你提供的代码进行训练的，冻结的只是前五十epoch的backbone，而上述所说的是neck中的结构  按理来说这w1w2w3w4是不会被冻结的。结算被冻结了50个epoch之后也会解冻的

 这是在冻结代码后设置断点debug后的结果，model中的w1w2w3w4的requires_grad都是=True

我将断点又设置到了loss.backward()后面，然后查看每迭代一次w1的变化，如下图所示发现w1的requires_gard=True，梯度gard也是一直有的，并且每次迭代后gard的值是在变化的，但是问题还是一样，w1的值没有更新一直都是【1,1】  下面是进一步迭代后的结果，可以看到gard在变化，但w1未更新 

最后发现应该是权重参数w不在optimizer.param_groups中，您有些代码中optimizer的初始化是optim.Adam(model.parameters()....),包含了model中所有的参数。但在目前我使用的这个代码中是进行筛选了的如下图  我知道的是这样筛选应该是剔除了我设定的权重参数w，但我没搞清楚您这样设置的目的是什么呢，为什么不直接用model.parameters()？

In issue2，‘Besides I visualized and found that the points are outside the grid model. Why is this happening.’ I forgot that i scaled the human mesh, and this is no...

> I tried a long time to predict the keypoints on a custom dataset, cleaned up the code, modified my depth images to be as close to the ITOP_side ones...

> I'm not clear about your experiments. > > 1. First I train it on H36M+MPII，the MPJPE is about 50+. This is normal -> what is the test set? >...

> I don't think this is from overfitting. I guess there is something wrong in your evaluation code? I checked the code and found that there was an issue with...

> Error 60 on training set means definitely underfitting Firstly, thank you for communicating with me. I know that continuing training may improve performance in the training set, but at...