maniache

Is this problem fixed now? I have the same issue and hope to download more pictures.

> 用Openpose生成的COCO关键点，需要删除Neck，才能兼容这个模型的关键点输入吧，我看COCO关键点是18个，而模型要求输入17个 您好，数据集repo提供的数据集地址我访问不了，请问您能帮我下载一份或是给我一份可以下载的链接吗？

@manansaxena @sanshibayuan @ankitsharma07 @Xuan-YE @asmallcodedog Hello everyone, I am sorry to bother you.But I can't access the dataset address provided by the dataset repo. Could you download it for me...

@liruilong940607 @manansaxena @wine3603 @marteiro Hello everyone, I am sorry to bother you.But I can't access the dataset address provided by the dataset repo. Could you download it for me or...

Hi, @anewell @ahangchen , I have the same doubts about this issue. Have you found out how to get the bounding box of a test image in MPII for correct...

Moreover, I want to confirm that, can I use the group border for testing? Could you share more details about your verification? This is deeply helpful to me. Thanks!

@xskxushaokai 我认为你的意思是全部关节点都从中心点回归； 不过论文提到由于长距离偏移比较难度可能较大，因此，作者将SPR扩展到了分层表示，子节点由父节点负责回归，也就是这里说的“父jioint的偏移值来学习子joint的位置”； 事实上objects as points 采用的应该是你所说的全部关节从中心点回归的策略，但是这篇文章作者也使用了原始的热图来提的关键点位置作为最终的结果，应该是也说明了长距离偏移不容易学习。

我也理解错了你的意思...不过你说的也很有道理，所说的两点好处应该是存在的； 但是我认为这样的偏移可能在另一种角度来看是比较难学的：我觉得在父节点放置子节点偏移响应的好处主要是在空间上编码了相对易学的躯干信息(事实上我觉得PAF有效的来源也就是编码了躯干信息)供网络学习，但是强制的将这样的响应搬到人体的中心位置可能会失去这样的“连续性”； 当然只是另一种角度的猜想，总之还是实验看看吧 (PS：受益很多，谢谢两位大佬)

Hey, @HansonXia I also found this problem when I was doing multi-person pose estimation. What solutions have you found out? Thanks in advance.

The same problem.