Error when any two cameras have different numbers of people

[letian-zhang]

Hi, I have run your code on the dataset Shelf. However, when the target camera and reference camera have different numbers of people, (e.g. there are 3 people in the target camera and there are 2 people in the reference camera), an error will happen in line 65 of mvmppe.py file.

poses_dist = torch.sum((pt - pr) ** 2, dim=-1) # [B, Npt, Npf, Nj, Nd]

Is there any solution to solve this problem?

[jiahaoLjh]

Hi,
Can you elaborate more on what error has happened?

[letian-zhang]

For example, there are two views at the same time.

The 500th frames of Camera0 and Camera1 have different numbers of people (e.g., 2 people in Camera0 and 4 people in Camaera1). In the code 'mvmppe.py' lines 63 and 64, the codes reshape the 2D target pose and 2D reference pose.

pt = poses_2d_target.reshape(batch_size, num_persons, 1, num_joints, num_depth_levels, 2) # [B, Npt, 1, Nj, Nd, 2]
pr = poses_2d_ref.reshape(batch_size, 1, num_persons, num_joints, 1, 2) # [B, 1, Npr, Nj, 1, 2]

The parameter 'num_persons' is the number of people in Camera0. However, there are 4 people in Camera1. When reshaping poses_2d_ref (Camera1) with parameter 'num_persons = 2', there is a error "RuntimeError: shape '[1, 1, 2, 17, 1, 2]' is invalid for input of size 136".

[jiahaoLjh]

The num_persons here is not the number of pose detections in each frame. It’s the maximum number of persons which you can see in the config files as “MAX_NUM_PERSONS”. Views with less than “MAX_NUM_PERSONS” pose detections will have dummy poses (all zeros) padded to the same shape.