camera parameters

[li1u]

Excuse me, I see that the camera parameters in your code are adjusted according to the size of the camera. I would like to ask if I want to use the generated camera parameters for reconstruction in other algorithms, not just visualization, what should I do?

[li1u]

The main problem is that the camera parameters obtained by other methods are completely wrong when reconstructed. My focal is obtained by estimate_focal_knowing_depth, and c2w is obtained by calibrate_camera_pnpransac.

`focals = estimate_focal_knowing_depth(pts3d[0][None].cuda(), valid_first[None].cuda()).cpu().item()

    intrinsics = torch.eye(3,)
    intrinsics[0, 0] = focals
    intrinsics[1, 1] = focals
    intrinsics[0, 2] = w / 2
    intrinsics[1, 2] = h / 2
    intrinsics = intrinsics.cuda()

    focals = torch.Tensor([focals]).reshape(1,).repeat(len(rgbimg))

    y_coords, x_coords = torch.meshgrid(torch.arange(h), torch.arange(w), indexing='ij')
    pixel_coords = torch.stack([x_coords, y_coords], dim=-1).float().cuda()  # [H, W, 2]

    c2ws = []
    for (pr_pt, valid) in zip(pts3d, msk):
        c2ws_i = calibrate_camera_pnpransac(pr_pt.cuda().flatten(0,1)[None], pixel_coords.flatten(0,1)[None], valid.cuda().flatten(0,1)[None], intrinsics[None])
        c2ws.append(c2ws_i[0])

    cams2world = torch.stack(c2ws, dim=0).cpu()  # [N, 4, 4]
    focals = to_numpy(focals)`

As shown in the picture, we actually shot around the object, but the effect shown is that the camera is in the center. Have you used the generated camera parameters?