Upload 2 files
Browse files- scripts/convert_to_lerobot.py +438 -0
- scripts/visualize_dataset.py +234 -0
scripts/convert_to_lerobot.py
ADDED
@@ -0,0 +1,438 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
This project is built upon the open-source project 🤗 LeRobot: https://github.com/huggingface/lerobot
|
3 |
+
|
4 |
+
We are grateful to the LeRobot team for their outstanding work and their contributions to the community.
|
5 |
+
|
6 |
+
If you find this project useful, please also consider supporting and exploring LeRobot.
|
7 |
+
"""
|
8 |
+
|
9 |
+
import os
|
10 |
+
import cv2
|
11 |
+
import json
|
12 |
+
import glob
|
13 |
+
import shutil
|
14 |
+
import logging
|
15 |
+
import argparse
|
16 |
+
from pathlib import Path
|
17 |
+
from typing import Callable
|
18 |
+
from functools import partial
|
19 |
+
from math import ceil
|
20 |
+
from copy import deepcopy
|
21 |
+
import subprocess
|
22 |
+
from multiprocessing import Pool, cpu_count
|
23 |
+
|
24 |
+
|
25 |
+
import h5py
|
26 |
+
import torch
|
27 |
+
import einops
|
28 |
+
import numpy as np
|
29 |
+
from PIL import Image
|
30 |
+
from tqdm import tqdm
|
31 |
+
|
32 |
+
|
33 |
+
HEAD_COLOR = "head.mp4"
|
34 |
+
HAND_LEFT_COLOR = "hand_left.mp4"
|
35 |
+
HAND_RIGHT_COLOR = "hand_right.mp4"
|
36 |
+
HEAD_CENTER_FISHEYE_COLOR = "head_front_fisheye.mp4"
|
37 |
+
HEAD_LEFT_FISHEYE_COLOR = "head_left_fisheye.mp4"
|
38 |
+
HEAD_RIGHT_FISHEYE_COLOR = "head_right_fisheye.mp4"
|
39 |
+
BACK_LEFT_FISHEYE_COLOR = "back_left_fisheye.mp4"
|
40 |
+
BACK_RIGHT_FISHEYE_COLOR = "back_right_fisheye.mp4"
|
41 |
+
HEAD_DEPTH = "head"
|
42 |
+
ALL_VIDEOS = [HEAD_COLOR, HAND_LEFT_COLOR, HAND_RIGHT_COLOR, HEAD_CENTER_FISHEYE_COLOR, HEAD_LEFT_FISHEYE_COLOR, HEAD_RIGHT_FISHEYE_COLOR, BACK_LEFT_FISHEYE_COLOR, BACK_RIGHT_FISHEYE_COLOR]
|
43 |
+
|
44 |
+
DEFAULT_IMAGE_PATH = (
|
45 |
+
"images/{image_key}/episode_{episode_index:06d}/frame_{frame_index:06d}.jpg"
|
46 |
+
)
|
47 |
+
|
48 |
+
FEATURES = {
|
49 |
+
"observation.images.top_head": {
|
50 |
+
"dtype": "video",
|
51 |
+
"shape": [480, 640, 3],
|
52 |
+
"names": ["height", "width", "channel"],
|
53 |
+
"video_info": {
|
54 |
+
"video.fps": 30.0,
|
55 |
+
"video.codec": "av1",
|
56 |
+
"video.pix_fmt": "yuv420p",
|
57 |
+
"video.is_depth_map": False,
|
58 |
+
"has_audio": False,
|
59 |
+
},
|
60 |
+
},
|
61 |
+
"observation.images.cam_top_depth": {
|
62 |
+
"dtype": "image",
|
63 |
+
"shape": [480, 640, 1],
|
64 |
+
"names": ["height", "width", "channel"],
|
65 |
+
},
|
66 |
+
"observation.images.hand_left": {
|
67 |
+
"dtype": "video",
|
68 |
+
"shape": [480, 640, 3],
|
69 |
+
"names": ["height", "width", "channel"],
|
70 |
+
"video_info": {
|
71 |
+
"video.fps": 30.0,
|
72 |
+
"video.codec": "av1",
|
73 |
+
"video.pix_fmt": "yuv420p",
|
74 |
+
"video.is_depth_map": False,
|
75 |
+
"has_audio": False,
|
76 |
+
},
|
77 |
+
},
|
78 |
+
"observation.images.hand_right": {
|
79 |
+
"dtype": "video",
|
80 |
+
"shape": [480, 640, 3],
|
81 |
+
"names": ["height", "width", "channel"],
|
82 |
+
"video_info": {
|
83 |
+
"video.fps": 30.0,
|
84 |
+
"video.codec": "av1",
|
85 |
+
"video.pix_fmt": "yuv420p",
|
86 |
+
"video.is_depth_map": False,
|
87 |
+
"has_audio": False,
|
88 |
+
},
|
89 |
+
},
|
90 |
+
"observation.images.head_center_fisheye": {
|
91 |
+
"dtype": "video",
|
92 |
+
"shape": [748, 960, 3],
|
93 |
+
"names": ["height", "width", "channel"],
|
94 |
+
"video_info": {
|
95 |
+
"video.fps": 30.0,
|
96 |
+
"video.codec": "av1",
|
97 |
+
"video.pix_fmt": "yuv420p",
|
98 |
+
"video.is_depth_map": False,
|
99 |
+
"has_audio": False,
|
100 |
+
},
|
101 |
+
},
|
102 |
+
"observation.images.head_left_fisheye": {
|
103 |
+
"dtype": "video",
|
104 |
+
"shape": [748, 960, 3],
|
105 |
+
"names": ["height", "width", "channel"],
|
106 |
+
"video_info": {
|
107 |
+
"video.fps": 30.0,
|
108 |
+
"video.codec": "av1",
|
109 |
+
"video.pix_fmt": "yuv420p",
|
110 |
+
"video.is_depth_map": False,
|
111 |
+
"has_audio": False,
|
112 |
+
},
|
113 |
+
},
|
114 |
+
"observation.images.head_right_fisheye": {
|
115 |
+
"dtype": "video",
|
116 |
+
"shape": [748, 960, 3],
|
117 |
+
"names": ["height", "width", "channel"],
|
118 |
+
"video_info": {
|
119 |
+
"video.fps": 30.0,
|
120 |
+
"video.codec": "av1",
|
121 |
+
"video.pix_fmt": "yuv420p",
|
122 |
+
"video.is_depth_map": False,
|
123 |
+
"has_audio": False,
|
124 |
+
},
|
125 |
+
},
|
126 |
+
"observation.images.back_left_fisheye": {
|
127 |
+
"dtype": "video",
|
128 |
+
"shape": [748, 960, 3],
|
129 |
+
"names": ["height", "width", "channel"],
|
130 |
+
"video_info": {
|
131 |
+
"video.fps": 30.0,
|
132 |
+
"video.codec": "av1",
|
133 |
+
"video.pix_fmt": "yuv420p",
|
134 |
+
"video.is_depth_map": False,
|
135 |
+
"has_audio": False,
|
136 |
+
},
|
137 |
+
},
|
138 |
+
"observation.images.back_right_fisheye": {
|
139 |
+
"dtype": "video",
|
140 |
+
"shape": [748, 960, 3],
|
141 |
+
"names": ["height", "width", "channel"],
|
142 |
+
"video_info": {
|
143 |
+
"video.fps": 30.0,
|
144 |
+
"video.codec": "av1",
|
145 |
+
"video.pix_fmt": "yuv420p",
|
146 |
+
"video.is_depth_map": False,
|
147 |
+
"has_audio": False,
|
148 |
+
},
|
149 |
+
},
|
150 |
+
"observation.state": {
|
151 |
+
"dtype": "float32",
|
152 |
+
"shape": [22],
|
153 |
+
},
|
154 |
+
"action": {
|
155 |
+
"dtype": "float32",
|
156 |
+
"shape": [22],
|
157 |
+
},
|
158 |
+
"episode_index": {
|
159 |
+
"dtype": "int64",
|
160 |
+
"shape": [1],
|
161 |
+
"names": None,
|
162 |
+
},
|
163 |
+
"frame_index": {
|
164 |
+
"dtype": "int64",
|
165 |
+
"shape": [1],
|
166 |
+
"names": None,
|
167 |
+
},
|
168 |
+
"index": {
|
169 |
+
"dtype": "int64",
|
170 |
+
"shape": [1],
|
171 |
+
"names": None,
|
172 |
+
},
|
173 |
+
"task_index": {
|
174 |
+
"dtype": "int64",
|
175 |
+
"shape": [1],
|
176 |
+
"names": None,
|
177 |
+
},
|
178 |
+
}
|
179 |
+
|
180 |
+
|
181 |
+
|
182 |
+
|
183 |
+
from modified_lerobot_dataset import AgiBotDataset
|
184 |
+
|
185 |
+
|
186 |
+
|
187 |
+
|
188 |
+
def process_video(video_path):
|
189 |
+
output = video_path.replace('.mp4', '_encode.mp4')
|
190 |
+
try:
|
191 |
+
command = [
|
192 |
+
"ffmpeg",
|
193 |
+
"-i", video_path,
|
194 |
+
"-vcodec", "libsvtav1",
|
195 |
+
"-pix_fmt", "yuv420p",
|
196 |
+
"-r", "30",
|
197 |
+
"-g", "2",
|
198 |
+
"-crf", "30",
|
199 |
+
"-vf", "scale=640:360:flags=bicubic",
|
200 |
+
"-loglevel", "error",
|
201 |
+
"-y", output
|
202 |
+
]
|
203 |
+
subprocess.run(command, check=True)
|
204 |
+
|
205 |
+
except subprocess.CalledProcessError as e:
|
206 |
+
print(f"Video failure: {' '.join(command)}, error: {e}")
|
207 |
+
except Exception as e:
|
208 |
+
print(f"Video unknwon failure: {' '.join(command)}, error: {e}")
|
209 |
+
finally:
|
210 |
+
pass
|
211 |
+
|
212 |
+
|
213 |
+
def preprocess_vides(episode_list, debug=False):
|
214 |
+
video_paths = []
|
215 |
+
for episode_path in episode_list:
|
216 |
+
video_dir = episode_path.replace('meta_info', 'observation') + "/video"
|
217 |
+
for file in ALL_VIDEOS:
|
218 |
+
video_path = os.path.join(video_dir, file)
|
219 |
+
video_paths.append(video_path)
|
220 |
+
|
221 |
+
if debug:
|
222 |
+
for video in video_paths:
|
223 |
+
process_video(video)
|
224 |
+
else:
|
225 |
+
with Pool(processes=os.cpu_count() // 2) as pool:
|
226 |
+
for _ in tqdm(pool.imap_unordered(process_video, video_paths), total=len(video_paths), desc="Video preprocessing"):
|
227 |
+
pass
|
228 |
+
|
229 |
+
|
230 |
+
def load_depths(root_dir: str, camera_name: str):
|
231 |
+
cam_path = Path(root_dir)
|
232 |
+
all_imgs = sorted(list(cam_path.glob(f"*")), key=lambda x: int(x.stem))
|
233 |
+
return [np.array(Image.open(f"{file}/{camera_name}.png")).astype(np.float32) / 1000 for file in all_imgs]
|
234 |
+
|
235 |
+
|
236 |
+
def load_local_dataset(episode_path: str) -> list | None:
|
237 |
+
"""Load local dataset and return a dict with observations and actions"""
|
238 |
+
observation_path = episode_path.replace('meta_info', 'observation')
|
239 |
+
with open(f"{episode_path}/task_info.json") as f:
|
240 |
+
task_info = json.load(f)
|
241 |
+
task = task_info['task_name']
|
242 |
+
|
243 |
+
with h5py.File(Path(episode_path) / "aligned_joints.h5") as f:
|
244 |
+
state_joint = np.array(f["state/joint/position"])
|
245 |
+
joint_names = f["state/joint"].attrs['name'].tolist()
|
246 |
+
|
247 |
+
|
248 |
+
head_joint_names = [
|
249 |
+
"joint_head_yaw",
|
250 |
+
"joint_head_pitch",
|
251 |
+
]
|
252 |
+
body_joint_names = [
|
253 |
+
"joint_lift_body",
|
254 |
+
"joint_body_pitch",
|
255 |
+
]
|
256 |
+
arm_joint_names = [
|
257 |
+
"Joint1_l",
|
258 |
+
"Joint1_r",
|
259 |
+
"Joint2_l",
|
260 |
+
"Joint2_r",
|
261 |
+
"Joint3_l",
|
262 |
+
"Joint3_r",
|
263 |
+
"Joint4_l",
|
264 |
+
"Joint4_r",
|
265 |
+
"Joint5_l",
|
266 |
+
"Joint5_r",
|
267 |
+
"Joint6_l",
|
268 |
+
"Joint6_r",
|
269 |
+
"Joint7_l",
|
270 |
+
"Joint7_r",
|
271 |
+
]
|
272 |
+
effector_joint_names = [
|
273 |
+
"right_Left_1_Joint",
|
274 |
+
"right_Right_1_Joint",
|
275 |
+
"left_Left_1_Joint",
|
276 |
+
"left_Right_1_Joint"
|
277 |
+
]
|
278 |
+
|
279 |
+
# Get indices for arm and effector joints from the first frame
|
280 |
+
head_joint_indices = [joint_names.index(name) for name in head_joint_names]
|
281 |
+
body_joint_indices = [joint_names.index(name) for name in body_joint_names]
|
282 |
+
arm_joint_indices = [joint_names.index(name) for name in arm_joint_names]
|
283 |
+
effector_joint_indices = [joint_names.index(name) for name in effector_joint_names]
|
284 |
+
|
285 |
+
# Extract joint positions for all frames
|
286 |
+
state_head = state_joint[:, head_joint_indices]
|
287 |
+
state_body = state_joint[:, body_joint_indices]
|
288 |
+
state_arm = state_joint[:, arm_joint_indices]
|
289 |
+
state_effector = state_joint[:, effector_joint_indices]
|
290 |
+
|
291 |
+
|
292 |
+
# Get action from state
|
293 |
+
action_head = state_head[1:] - state_head[:-1]
|
294 |
+
action_body = state_body[1:] - state_body[:-1]
|
295 |
+
action_arm = state_arm[1:] - state_arm[:-1]
|
296 |
+
action_effector = state_effector[1:] - state_effector[:-1]
|
297 |
+
|
298 |
+
# repeat the last frame of the action
|
299 |
+
action_head = np.concatenate([action_head, action_head[-1:]])
|
300 |
+
action_body = np.concatenate([action_body, action_body[-1:]])
|
301 |
+
action_arm = np.concatenate([action_arm, action_arm[-1:]])
|
302 |
+
action_effector = np.concatenate([action_effector, action_effector[-1:]])
|
303 |
+
|
304 |
+
|
305 |
+
|
306 |
+
states_value = np.hstack(
|
307 |
+
[state_head, state_body, state_arm, state_effector]
|
308 |
+
).astype(np.float32)
|
309 |
+
assert (
|
310 |
+
action_arm.shape[0] == action_effector.shape[0]
|
311 |
+
), f"shape of action_arm:{action_arm.shape};shape of action_effector:{action_effector.shape}"
|
312 |
+
action_value = np.hstack(
|
313 |
+
[action_head, action_body, action_arm, action_effector]
|
314 |
+
).astype(np.float32)
|
315 |
+
|
316 |
+
depth_imgs = load_depths(f"{observation_path}/depth", HEAD_DEPTH)
|
317 |
+
|
318 |
+
assert len(depth_imgs) == len(
|
319 |
+
states_value
|
320 |
+
), f"Number of images and states are not equal"
|
321 |
+
assert len(depth_imgs) == len(
|
322 |
+
action_value
|
323 |
+
), f"Number of images and actions are not equal"
|
324 |
+
frames = [
|
325 |
+
{
|
326 |
+
"observation.images.cam_top_depth": depth_imgs[i],
|
327 |
+
"observation.state": states_value[i],
|
328 |
+
"action": action_value[i],
|
329 |
+
}
|
330 |
+
for i in range(len(depth_imgs))
|
331 |
+
]
|
332 |
+
|
333 |
+
v_path = observation_path + "/video"
|
334 |
+
videos = {
|
335 |
+
"observation.images.top_head": f"{v_path}/{HEAD_COLOR}".replace('.mp4', '_encode.mp4'),
|
336 |
+
"observation.images.hand_left": f"{v_path}/{HAND_LEFT_COLOR}".replace('.mp4', '_encode.mp4'),
|
337 |
+
"observation.images.hand_right": f"{v_path}/{HAND_RIGHT_COLOR}".replace('.mp4', '_encode.mp4'),
|
338 |
+
"observation.images.head_center_fisheye": f"{v_path}/{HEAD_CENTER_FISHEYE_COLOR}".replace('.mp4', '_encode.mp4'),
|
339 |
+
"observation.images.head_left_fisheye": f"{v_path}/{HEAD_LEFT_FISHEYE_COLOR}".replace('.mp4', '_encode.mp4'),
|
340 |
+
"observation.images.head_right_fisheye": f"{v_path}/{HEAD_RIGHT_FISHEYE_COLOR}".replace('.mp4', '_encode.mp4'),
|
341 |
+
"observation.images.back_left_fisheye": f"{v_path}/{BACK_LEFT_FISHEYE_COLOR}".replace('.mp4', '_encode.mp4'),
|
342 |
+
"observation.images.back_right_fisheye": f"{v_path}/{BACK_RIGHT_FISHEYE_COLOR}".replace('.mp4', '_encode.mp4'),
|
343 |
+
}
|
344 |
+
return {
|
345 |
+
'frames': frames,
|
346 |
+
'videos': videos,
|
347 |
+
'task': task
|
348 |
+
}
|
349 |
+
|
350 |
+
|
351 |
+
|
352 |
+
|
353 |
+
|
354 |
+
def main(
|
355 |
+
src_path: str,
|
356 |
+
tgt_path: str,
|
357 |
+
repo_id: str,
|
358 |
+
preprocess_video: bool = False,
|
359 |
+
debug: bool = True,
|
360 |
+
):
|
361 |
+
# remove the existing dataset
|
362 |
+
if os.path.exists(f"{tgt_path}/{repo_id}"):
|
363 |
+
shutil.rmtree(f"{tgt_path}/{repo_id}")
|
364 |
+
dataset = AgiBotDataset.create(
|
365 |
+
repo_id=repo_id,
|
366 |
+
root=f"{tgt_path}/{repo_id}",
|
367 |
+
fps=30,
|
368 |
+
robot_type="a2d",
|
369 |
+
features=FEATURES,
|
370 |
+
)
|
371 |
+
|
372 |
+
episode_list = sorted(
|
373 |
+
[
|
374 |
+
f
|
375 |
+
for f in glob.glob(f"{src_path}/meta_info/*/*")
|
376 |
+
if os.path.isdir(f)
|
377 |
+
]
|
378 |
+
)
|
379 |
+
|
380 |
+
# preprocess the videos to avoid encoding error
|
381 |
+
if preprocess_video:
|
382 |
+
preprocess_vides(episode_list, debug)
|
383 |
+
|
384 |
+
# load the raw datasets
|
385 |
+
raw_datasets_before_filter = [
|
386 |
+
load_local_dataset(episode_path)
|
387 |
+
for episode_path in tqdm(episode_list)
|
388 |
+
]
|
389 |
+
|
390 |
+
# remove the None result from the raw_datasets
|
391 |
+
raw_datasets = [
|
392 |
+
dataset for dataset in raw_datasets_before_filter if dataset is not None
|
393 |
+
]
|
394 |
+
|
395 |
+
for episode_data in tqdm(raw_datasets, desc="Generating dataset from raw datasets"):
|
396 |
+
for frame in tqdm(episode_data['frames'], desc="Generating dataset from raw dataset"):
|
397 |
+
dataset.add_frame(frame)
|
398 |
+
|
399 |
+
dataset.save_episode(task=episode_data['task'], videos=episode_data['videos'])
|
400 |
+
dataset.consolidate(run_compute_stats=True)
|
401 |
+
|
402 |
+
|
403 |
+
|
404 |
+
|
405 |
+
|
406 |
+
if __name__ == "__main__":
|
407 |
+
parser = argparse.ArgumentParser()
|
408 |
+
parser.add_argument(
|
409 |
+
"--data_dir",
|
410 |
+
type=str,
|
411 |
+
required=True,
|
412 |
+
)
|
413 |
+
parser.add_argument(
|
414 |
+
"--save_dir",
|
415 |
+
type=str,
|
416 |
+
required=True,
|
417 |
+
)
|
418 |
+
parser.add_argument(
|
419 |
+
"--repo_id",
|
420 |
+
type=str,
|
421 |
+
required=True,
|
422 |
+
)
|
423 |
+
parser.add_argument(
|
424 |
+
"--preprocess_video",
|
425 |
+
action="store_true",
|
426 |
+
)
|
427 |
+
parser.add_argument(
|
428 |
+
"--debug",
|
429 |
+
action="store_true",
|
430 |
+
)
|
431 |
+
args = parser.parse_args()
|
432 |
+
|
433 |
+
|
434 |
+
|
435 |
+
assert os.path.exists(args.data_dir), f"Cannot find {args.data_dir}."
|
436 |
+
|
437 |
+
|
438 |
+
main(args.data_dir, args.save_dir, args.repo_id, args.preprocess_video, args.debug)
|
scripts/visualize_dataset.py
ADDED
@@ -0,0 +1,234 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
This script is adapted from the Hugging Face 🤗 LeRobot project:
|
3 |
+
https://github.com/huggingface/lerobot
|
4 |
+
|
5 |
+
Original file:
|
6 |
+
https://github.com/huggingface/lerobot/blob/main/lerobot/scripts/visualize_dataset.py
|
7 |
+
|
8 |
+
The original script was developed as part of the LeRobot project for dataset visualization.
|
9 |
+
This version adds support for depth map visualization.
|
10 |
+
"""
|
11 |
+
|
12 |
+
import argparse
|
13 |
+
import gc
|
14 |
+
import logging
|
15 |
+
import time
|
16 |
+
from pathlib import Path
|
17 |
+
from typing import Iterator
|
18 |
+
|
19 |
+
import numpy as np
|
20 |
+
import rerun as rr
|
21 |
+
import torch
|
22 |
+
import torch.utils.data
|
23 |
+
import tqdm
|
24 |
+
import matplotlib.pyplot as plt
|
25 |
+
|
26 |
+
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
|
27 |
+
|
28 |
+
|
29 |
+
class EpisodeSampler(torch.utils.data.Sampler):
|
30 |
+
def __init__(self, dataset: LeRobotDataset, episode_index: int):
|
31 |
+
from_idx = dataset.episode_data_index["from"][episode_index].item()
|
32 |
+
to_idx = dataset.episode_data_index["to"][episode_index].item()
|
33 |
+
self.frame_ids = range(from_idx, to_idx)
|
34 |
+
|
35 |
+
def __iter__(self) -> Iterator:
|
36 |
+
return iter(self.frame_ids)
|
37 |
+
|
38 |
+
def __len__(self) -> int:
|
39 |
+
return len(self.frame_ids)
|
40 |
+
|
41 |
+
|
42 |
+
def to_hwc_uint8_numpy(chw_float32_torch: torch.Tensor) -> np.ndarray:
|
43 |
+
assert chw_float32_torch.dtype == torch.float32
|
44 |
+
assert chw_float32_torch.ndim == 3
|
45 |
+
c, h, w = chw_float32_torch.shape
|
46 |
+
assert c < h and c < w, f"Expect channel first images, but instead {chw_float32_torch.shape}"
|
47 |
+
|
48 |
+
if c == 1:
|
49 |
+
# If depth image, clip and normalize the depth map just for visualization
|
50 |
+
min_depth = 0.4
|
51 |
+
max_depth = 3
|
52 |
+
clipped_depth = torch.clamp(chw_float32_torch, min=min_depth, max=max_depth)
|
53 |
+
normalized_depth = (clipped_depth-min_depth) / (max_depth-min_depth)
|
54 |
+
depth_image = np.sqrt(normalized_depth.squeeze().cpu().numpy())
|
55 |
+
|
56 |
+
colormap = plt.get_cmap('jet')
|
57 |
+
colored_depth_image = colormap(depth_image)
|
58 |
+
hwc_uint8_numpy = (colored_depth_image[:, :, :3] * 255).astype(np.uint8)
|
59 |
+
else:
|
60 |
+
# If RGB image
|
61 |
+
hwc_uint8_numpy = (chw_float32_torch * 255).type(torch.uint8).permute(1, 2, 0).numpy()
|
62 |
+
|
63 |
+
return hwc_uint8_numpy
|
64 |
+
|
65 |
+
|
66 |
+
def visualize_dataset(
|
67 |
+
dataset: LeRobotDataset,
|
68 |
+
episode_index: int,
|
69 |
+
batch_size: int = 32,
|
70 |
+
num_workers: int = 0,
|
71 |
+
mode: str = "local",
|
72 |
+
web_port: int = 9090,
|
73 |
+
ws_port: int = 9087,
|
74 |
+
save: bool = False,
|
75 |
+
output_dir: Path | None = None,
|
76 |
+
**kwargs,
|
77 |
+
) -> Path | None:
|
78 |
+
if save:
|
79 |
+
assert (
|
80 |
+
output_dir is not None
|
81 |
+
), "Set an output directory where to write .rrd files with `--output-dir path/to/directory`."
|
82 |
+
|
83 |
+
repo_id = dataset.repo_id
|
84 |
+
|
85 |
+
logging.info("Loading dataloader")
|
86 |
+
episode_sampler = EpisodeSampler(dataset, episode_index)
|
87 |
+
dataloader = torch.utils.data.DataLoader(
|
88 |
+
dataset,
|
89 |
+
num_workers=num_workers,
|
90 |
+
batch_size=batch_size,
|
91 |
+
sampler=episode_sampler,
|
92 |
+
)
|
93 |
+
|
94 |
+
logging.info("Starting Rerun")
|
95 |
+
|
96 |
+
if mode not in ["local", "distant"]:
|
97 |
+
raise ValueError(mode)
|
98 |
+
|
99 |
+
spawn_local_viewer = mode == "local" and not save
|
100 |
+
rr.init(f"{repo_id}/episode_{episode_index}", spawn=spawn_local_viewer)
|
101 |
+
|
102 |
+
# Manually call python garbage collector after `rr.init` to avoid hanging in a blocking flush
|
103 |
+
# when iterating on a dataloader with `num_workers` > 0
|
104 |
+
# TODO(rcadene): remove `gc.collect` when rerun version 0.16 is out, which includes a fix
|
105 |
+
gc.collect()
|
106 |
+
|
107 |
+
if mode == "distant":
|
108 |
+
rr.serve(open_browser=False, web_port=web_port, ws_port=ws_port)
|
109 |
+
|
110 |
+
logging.info("Logging to Rerun")
|
111 |
+
|
112 |
+
for batch in tqdm.tqdm(dataloader, total=len(dataloader)):
|
113 |
+
# iterate over the batch
|
114 |
+
for i in range(len(batch["index"])):
|
115 |
+
rr.set_time_sequence("frame_index", batch["frame_index"][i].item())
|
116 |
+
rr.set_time_seconds("timestamp", batch["timestamp"][i].item())
|
117 |
+
|
118 |
+
# display each camera image
|
119 |
+
for key in dataset.meta.camera_keys:
|
120 |
+
# TODO(rcadene): add `.compress()`? is it lossless?
|
121 |
+
rr.log(key, rr.Image(to_hwc_uint8_numpy(batch[key][i])))
|
122 |
+
|
123 |
+
# display each dimension of action space (e.g. actuators command)
|
124 |
+
if "action" in batch:
|
125 |
+
for dim_idx, val in enumerate(batch["action"][i]):
|
126 |
+
rr.log(f"action/{dim_idx}", rr.Scalar(val.item()))
|
127 |
+
|
128 |
+
# display each dimension of observed state space (e.g. agent position in joint space)
|
129 |
+
if "observation.state" in batch:
|
130 |
+
for dim_idx, val in enumerate(batch["observation.state"][i]):
|
131 |
+
rr.log(f"state/{dim_idx}", rr.Scalar(val.item()))
|
132 |
+
|
133 |
+
if mode == "local" and save:
|
134 |
+
# save .rrd locally
|
135 |
+
output_dir = Path(output_dir)
|
136 |
+
output_dir.mkdir(parents=True, exist_ok=True)
|
137 |
+
repo_id_str = repo_id.replace("/", "_")
|
138 |
+
rrd_path = output_dir / f"{repo_id_str}_episode_{episode_index}.rrd"
|
139 |
+
rr.save(rrd_path)
|
140 |
+
return rrd_path
|
141 |
+
|
142 |
+
elif mode == "distant":
|
143 |
+
# stop the process from exiting since it is serving the websocket connection
|
144 |
+
try:
|
145 |
+
while True:
|
146 |
+
time.sleep(1)
|
147 |
+
except KeyboardInterrupt:
|
148 |
+
print("Ctrl-C received. Exiting.")
|
149 |
+
|
150 |
+
|
151 |
+
def main():
|
152 |
+
parser = argparse.ArgumentParser()
|
153 |
+
|
154 |
+
parser.add_argument(
|
155 |
+
"--repo-id",
|
156 |
+
type=str,
|
157 |
+
default=None,
|
158 |
+
)
|
159 |
+
parser.add_argument(
|
160 |
+
"--episode-index",
|
161 |
+
type=int,
|
162 |
+
nargs="*",
|
163 |
+
default=None,
|
164 |
+
help="Episode indices to visualize (e.g. `0 1 5 6` to load episodes of index 0, 1, 5 and 6). By default loads all episodes.",
|
165 |
+
)
|
166 |
+
parser.add_argument(
|
167 |
+
"--dataset-path",
|
168 |
+
type=Path,
|
169 |
+
default=None,
|
170 |
+
help="Root directory for the converted LeRobot dataset stored locally.",
|
171 |
+
)
|
172 |
+
parser.add_argument(
|
173 |
+
"--output-dir",
|
174 |
+
type=Path,
|
175 |
+
default=None,
|
176 |
+
help="Directory path to write a .rrd file when `--save 1` is set.",
|
177 |
+
)
|
178 |
+
parser.add_argument(
|
179 |
+
"--batch-size",
|
180 |
+
type=int,
|
181 |
+
default=32,
|
182 |
+
help="Batch size loaded by DataLoader.",
|
183 |
+
)
|
184 |
+
parser.add_argument(
|
185 |
+
"--num-workers",
|
186 |
+
type=int,
|
187 |
+
default=4,
|
188 |
+
help="Number of processes of Dataloader for loading the data.",
|
189 |
+
)
|
190 |
+
parser.add_argument(
|
191 |
+
"--mode",
|
192 |
+
type=str,
|
193 |
+
default="local",
|
194 |
+
help=(
|
195 |
+
"Mode of viewing between 'local' or 'distant'. "
|
196 |
+
"'local' requires data to be on a local machine. It spawns a viewer to visualize the data locally. "
|
197 |
+
"'distant' creates a server on the distant machine where the data is stored. "
|
198 |
+
"Visualize the data by connecting to the server with `rerun ws://localhost:PORT` on the local machine."
|
199 |
+
),
|
200 |
+
)
|
201 |
+
parser.add_argument(
|
202 |
+
"--web-port",
|
203 |
+
type=int,
|
204 |
+
default=9090,
|
205 |
+
help="Web port for rerun.io when `--mode distant` is set.",
|
206 |
+
)
|
207 |
+
parser.add_argument(
|
208 |
+
"--ws-port",
|
209 |
+
type=int,
|
210 |
+
default=9087,
|
211 |
+
help="Web socket port for rerun.io when `--mode distant` is set.",
|
212 |
+
)
|
213 |
+
parser.add_argument(
|
214 |
+
"--save",
|
215 |
+
type=int,
|
216 |
+
default=0,
|
217 |
+
help=(
|
218 |
+
"Save a .rrd file in the directory provided by `--output-dir`. "
|
219 |
+
"It also deactivates the spawning of a viewer. "
|
220 |
+
"Visualize the data by running `rerun path/to/file.rrd` on your local machine."
|
221 |
+
),
|
222 |
+
)
|
223 |
+
|
224 |
+
args = parser.parse_args()
|
225 |
+
kwargs = vars(args)
|
226 |
+
root = f"{kwargs.pop('dataset_path')}/{args.repo_id}"
|
227 |
+
|
228 |
+
logging.info("Loading dataset")
|
229 |
+
dataset = LeRobotDataset(args.repo_id, root=root, local_files_only=True)
|
230 |
+
|
231 |
+
visualize_dataset(dataset, **vars(args))
|
232 |
+
|
233 |
+
if __name__ == "__main__":
|
234 |
+
main()
|