Question about getting single person data from custom video

[QMME]

Hello! I have recently run your work. I want to get one main person from my video. However there are some other persons showing up during the video sometimes. How should i set the configs? I have tried set "multi_person" to False, but seems it won't work.
It would be so nice if you could help me here. Thank you.

[Arthur151]

I think you need to track a specific person through the video. Sorry, the current version code doesn't support this function. The beta version that has this tracking function is still under development. I will update this issue when I upload the beta version.
If the video scene is simple, you can try to use the center location to track the specific person and filter out the other people. Please feel free to contact me if there is any problem in this process.

[QMME]

Thank you so much and i expect the updating.

[bedman367]

我也有同样的问题，比如表演者背对着镜面墙体，墙内时不时出现的反光镜像的姿态信息也会被提取出来。但是提取出来的姿态数据还是（frame,72）的形式，似乎是把表演者和影子的姿态信息融合到一起了？ 请问除了对原视频进行手动处理目前没有其他更好的解决方法了嘛？期待您的回复。

[Arthur151]

您好，目前的simple-romp是支持基础的tracking的，可以提取到动作序列。

[bedman367]

谢谢您的回复！但是目前我们是基于之前的romp提取的姿态信息，任务比较急，换simple_romp可能需要把已经提取好的视频全部重新提一遍，时间成本比较大（当然主要是因为本人水平有限23333）
关于代码的细节不知可否咨询一下？
我看返回结果时都会调用reorganize函数组织结果格式，其中这里遍历subject_idx，这里是不是就对一个检测到的每个表演者啊？ 如果是的话我改一下保存的代码，只保留subject_idx=0的数据，是不是就能解决我的问题了？

[Arthur151]

@bedman367
如果你可以确定只想要图像中估计的尺度最大的人的姿态的话，可以根据估计的cam信息

ROMP/romp/predict/base_predictor.py

Line 71 in e4613fd

results[img_path][subject_idx]['cam'] = cam_results[batch_idx]

cam是3-dim的，第0个就是scale信息，可以据此获得尺度最大人的index，然后把这个人的结果提取出来。

具体代码写法可以参考：

ROMP/simple_romp/romp/main.py

Line 130 in e4613fd

max_id = torch.argmax(outputs['cam'][:,0])

[bedman367]

所以subject idx就是可以理解为检测到的每个人的id，我只需找出尺度最大的那个id就可以了，多谢！

[Arthur151]

太客气了，如果有其他需要讨论的，请单独开issue，我差点漏掉了这个。