inverse comment fix in pred_extrinsic process_ray_pose_estimation

[Golbstein]

we actually return w2c as extrinsic

[jianglh-WHU]

I think the author might have reversed the direction and origin in the paper.

Because ⬇️ Obviously, direction is in the first three channels.

Depth-Anything-3/src/depth_anything_3/utils/ray_utils.py

Lines 484 to 493 in 81083bb

	R, focal_lengths, principal_points = compute_optimal_rotation_intrinsics_batch(
	I_cam_plane_unproj,
	camray[:, :, :3],
	reproj_threshold=reproj_threshold,
	weights=confidence,
	n_sample = n_sample,
	n_iter=n_iter,
	num_sample_for_ransac=num_sample_for_ransac,
	rand_sample_iters_idx=rand_sample_iters_idx,
	)

[Golbstein]

yes, they did. but that's a PR for their paper ;)

[jianglh-WHU]

hh, just discuss.