About the generalization capability of the model

[logic03]

May I use the weight of ITOP data set to detect the key points of the depth map collected by my camera?
I found that there is mean and std value in the ITOP data set. Does this step also need to be carried out in the depth map I collected by myself?

---

请问我可以使用ITOP数据集的权重去检测自己的相机采集到的深度图的关键点吗？我发现ITOP数据集里面有一个mean和std，我自己采集的深度图也需要进行这样一步计算吗？还是说ITOP这个数据集采集的数据无法泛化到现实情况中呢？

[mks0601]

I found that there is mean and std value in the ITOP data set. <- what do you mean by the mean and std? Are they defined in my codes? If so, could you let me know where they are defined?

[logic03]

I found that there is mean and std value in the ITOP data set. <- what do you mean by the mean and std? Are they defined in my codes? If so, could you let me know where they are defined?

对不起，其实我是在复现这个iccv2019代码的结果，他也是用到了ITOP数据集。
https://github.com/zhangboshen/A2J
他的网络是直接输入2D图像的深度信息，经过归一化，直接分支回归得出ITOP所标注的3D坐标。
我认为将我自己用Azure kinect采集得到的深度图（以m为单位的深度值）传入这个网络，使用他在ITOP数据集上预训练的权重推断应该能得出合理的结果。
但是我尝试了发现推断出来的15个关节点的结果不正确。
我发现他的代码里面计算了一个所有图像的mean和std标准差，相当于就是在原深度图像像素（以m为单位的深度值）上进行了归一化，这样改变了原深度图像的深度信息。
我无法理解这个操作，看到您的论文也用到了这个数据集以为您的也是使用了这两个数值，所以想向您请教一下为什么要进行这个步骤。
zhangboshen/A2J#12

---

Sorry, I'm actually replicating the result of this ICCV2019 code, which also uses the ITOP data set.
https://github.com/zhangboshen/A2J
(1)In his network, the depth information of 2D images was directly input, and the 3D coordinates marked by ITOP were obtained by direct branch regression after normalization.
(2)I think it would be reasonable to pass my own depth map (depth value in m units) collected with Azure Kinect into this network and infer the weights pre-trained by him on ITOP data sets.
(3)But I tried and found that the inferred results of the 15 nodes were not correct.
(4)I found that his code calculated a mean and STD standard deviation of all images, which was equivalent to normalizing the original depth image pixels (depth value in m units), thus changing the depth information of the original depth image.
(5)I can't understand this operation. When I saw that this data set was also used in your paper, I assumed that these two values were also used in yours, so I would like to ask you why I did this step.
zhangboshen/A2J#12

[mks0601]

You'd better ask this to the authors of that paper, not me.

[justintiger]

May I use the weight of ITOP data set to detect the key points of the depth map collected by my camera?
I found that there is mean and std value in the ITOP data set. Does this step also need to be carried out in the depth map I collected by myself?

请问我可以使用ITOP数据集的权重去检测自己的相机采集到的深度图的关键点吗？我发现ITOP数据集里面有一个mean和std，我自己采集的深度图也需要进行这样一步计算吗？还是说ITOP这个数据集采集的数据无法泛化到现实情况中呢？

I also do the same project like u. Did u solve thie problem? Can we maybe keep in touch in wechat? 385334833