|
|
|
|
|
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 |
|
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 |
