Method get_pose_rank_ids

data/recon_dataset.py:377–419

(self, total_ids, extrinsics, expand_ratio=None, upper_bound=None)

        return result_ids
        
    def get_pose_rank_ids(self, total_ids, extrinsics, expand_ratio=None, upper_bound=None):
    
        num_views = len(extrinsics)
        ranking, dists = compute_ranking(extrinsics, lambda_t=1.0, normalize=True, batched=True)
        # reference_view = random.sample(range(num_views), 1)[0]
        reference_view = np.random.randint(0, num_views)
        refview_ranking = ranking[reference_view]

        start_idx = 0

        # Determine the actual expand_range
        if upper_bound is None:
            # Use ratio to determine range
            expand_range = int(total_ids * expand_ratio * 2)
        else:
            expand_range = min(upper_bound, int(total_ids * expand_ratio * 2))
        high_bound = min(num_views, start_idx + expand_range)

        # Create the valid range of indices
        valid_ranking_ids = np.arange(start_idx, high_bound)
        valid_range = refview_ranking[valid_ranking_ids]

        # Sample 'total_ids - 1' items, because we already have the start_idx
        sampled_ids = np.random.choice(
            valid_range,
            size=(total_ids - 1),
            replace=True,   # we accept the situation that some sampled ids are the same
        )

        # start_randk_element = refview_ranking[start_idx]
        # return np.insert(sampled_ids, 0, start_randk_element)

        # # for nvs, farthest views must be at the start and end of the sequence
        # sampled_rankings = [np.where(refview_ranking == id_val)[0][0] for id_val in sampled_ids]
        # farthest_rank_idx = np.argmax(sampled_rankings)  # Find the element with smallest ranking (highest rank)

        # farthest_rank_element = sampled_ids[farthest_rank_idx]
        # sampled_ids = np.delete(sampled_ids, farthest_rank_idx)
        # sampled_ids = np.append(sampled_ids, farthest_rank_element)

        start_randk_element = refview_ranking[start_idx]
        result_ids = np.insert(sampled_ids, 0, start_randk_element)
        return result_ids