Spaces:
Building
Building
| # Copyright (c) Meta Platforms, Inc. and affiliates. | |
| import numpy as np | |
| import torch | |
| def chunk_sequence( | |
| data, | |
| indices, | |
| *, | |
| names=None, | |
| max_length=100, | |
| min_length=1, | |
| max_delay_s=None, | |
| max_inter_dist=None, | |
| max_total_dist=None, | |
| ): | |
| sort_array = data.get("capture_time", data.get("index")) | |
| if sort_array is None: | |
| sort_array = indices if names is None else names | |
| indices = sorted(indices, key=lambda i: sort_array[i].tolist()) | |
| centers = torch.stack([data["t_c2w"][i][:2] for i in indices]).numpy() | |
| dists = np.linalg.norm(np.diff(centers, axis=0), axis=-1) | |
| if "capture_time" in data: | |
| times = torch.stack([data["capture_time"][i] for i in indices]) | |
| times = times.double() / 1e3 # ms to s | |
| delays = np.diff(times, axis=0) | |
| else: | |
| delays = np.zeros_like(dists) | |
| chunks = [[indices[0]]] | |
| dist_total = 0 | |
| for dist, delay, idx in zip(dists, delays, indices[1:]): | |
| dist_total += dist | |
| if ( | |
| (max_inter_dist is not None and dist > max_inter_dist) | |
| or (max_total_dist is not None and dist_total > max_total_dist) | |
| or (max_delay_s is not None and delay > max_delay_s) | |
| or len(chunks[-1]) >= max_length | |
| ): | |
| chunks.append([]) | |
| dist_total = 0 | |
| chunks[-1].append(idx) | |
| chunks = list(filter(lambda c: len(c) >= min_length, chunks)) | |
| chunks = sorted(chunks, key=len, reverse=True) | |
| return chunks |