Value error

[Zumbalamambo]

I tried depth_estimation.ipynb.

# warp by depth
i_trans_j = i_pose_w @ torch.inverse(j_pose_w)
k_trans_j = k_pose_w @ torch.inverse(j_pose_w)
dst_trans_src = torch.cat([i_trans_j, k_trans_j], dim=0)

depth_src = depth_j
intrinsics_src = intrinsics_j

image_src = image_j
image_dst = torch.cat([image_i, image_k], dim=0)

image_dst_to_src = kornia.warp_frame_depth(image_dst, depth_src, dst_trans_src, intrinsics_src)
print(image_dst_to_src.shape)

It throws the following error,

ValueError: Input batch size must be the same for both tensors or 1

[edgarriba]

@Zumbalamambo what are the actual shapes of your inputs ?

[Zumbalamambo]

@edgarriba torch.Size([1, 3, 240, 320]) is the actual size.

I got the same error on my other computer too. I just cloned the repo and tried this without varying any single parameters or files.

[Zumbalamambo]

print(image_i.shape) -> torch.Size([1, 3, 240, 320])
print(image_j.shape) -> torch.Size([1, 3, 240, 320])
print(image_k.shape) -> torch.Size([1, 3, 240, 320])

print(image_dst.shape) -> torch.Size([2, 3, 240, 320])
print(depth_src.shape) -> torch.Size([1, 1, 240, 320])
print(dst_trans_src.shape) -> torch.Size([2, 4, 4])
print(intrinsics_src.shape) -> torch.Size([1, 3, 3])

[edgarriba]

@Zumbalamambo you batch size should be equal for every input tensor, need to double check that broadcasting works here.

[Zumbalamambo]

@Zumbalamambo I'm not sure what changes the batch size. I did not do anything different. Its the fresh clone

[edgarriba]

@Zumbalamambo What I mean is that your depth_src should be as same as image_dst and also intrinsics_src.
Try this: depth_src = depth_src.expand(image_dst.shape[0], -1, -1, -1)

[Zumbalamambo]

@edgarriba that works but it throws the following error in the last cell,

`---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-13-76c8d1a0dfb0> in <module>
     26     compute_scale_loss(image_dst_scale, image_src_scale, depth_src_pred,
     27                        dst_trans_src, intrinsics_src_scale, optimizer,
---> 28                        num_iterations, error_tol)
     29 
     30     print('Train iteration: {}/{}'.format(iter_idx, num_levels))

<ipython-input-12-14f8071ed76a> in compute_scale_loss(image_dst, image_src, depth_src, dst_trans_src, intrinsics_src, optimizer, num_iterations, error_tol)
     15 
     16             image_dst_to_src = kornia.warp_frame_depth(
---> 17                 image_dst, depth_src_tmp, dst_trans_src, intrinsics_src)
     18 
     19             ones = kornia.warp_frame_depth(torch.ones_like(image_dst),

~/anaconda3/envs/slam/lib/python3.6/site-packages/kornia/geometry/depth.py in warp_frame_depth(image_src, depth_dst, src_trans_dst, camera_matrix, normalize_points)
    163 
    164     # apply transformation to the 3d points
--> 165     points_3d_src = transform_points(src_trans_dst[:, None], points_3d_dst)  # BxHxWx3
    166 
    167     # project back to pixels

~/anaconda3/envs/slam/lib/python3.6/site-packages/kornia/geometry/linalg.py in transform_points(trans_01, points_1)
    203         raise TypeError("Tensor must be in the same device")
    204     if not trans_01.shape[0] == points_1.shape[0] and trans_01.shape[0] != 1:
--> 205         raise ValueError("Input batch size must be the same for both tensors or 1")
    206     if not trans_01.shape[-1] == (points_1.shape[-1] + 1):
    207         raise ValueError("Last input dimensions must differe by one unit")

ValueError: Input batch size must be the same for both tensors or 1
`

The error points to

[edgarriba]

@Zumbalamambo you need to do the same for the intrinsics matrix. See: https://colab.research.google.com/drive/1DX4l_ssSEXSofOR3Hm9TcgRwFDWVEvJO?usp=sharing

[Zumbalamambo]

@edgarriba unfortunately the colab file is protected :(

[edgarriba]

@Zumbalamambo try again. Let me know if it works and if we should update the original example.

[Zumbalamambo]

@edgarriba that works but still the last cell of the notebook throws the error that I have mentioned above.