File size: 5,673 Bytes
15fd275 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 | #!/usr/bin/env python3
"""Render the three RGB views, exact uint16 depth, and both tactile surfaces."""
from __future__ import annotations
import argparse
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
RGB_KEYS = (
("observation.images.cam_front", "Front RGB"),
("observation.images.cam_side", "Side RGB"),
("observation.images.cam_fisheye", "Fisheye RGB"),
)
def lerobot_dataset_class():
try:
from lerobot.datasets.lerobot_dataset import LeRobotDataset
except ModuleNotFoundError:
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
return LeRobotDataset
def to_numpy(value: object) -> np.ndarray:
if hasattr(value, "detach"):
value = value.detach().cpu().numpy()
return np.asarray(value)
def rgb_hwc(value: object) -> np.ndarray:
image = to_numpy(value)
if image.ndim != 3:
raise ValueError(f"expected RGB rank 3, got {image.shape}")
if image.shape[0] in (1, 3, 4) and image.shape[-1] not in (1, 3, 4):
image = np.moveaxis(image, 0, -1)
if image.shape[-1] == 4:
image = image[..., :3]
if np.issubdtype(image.dtype, np.integer):
image = image.astype(np.float32) / 255.0
return np.clip(image, 0.0, 1.0)
def exact_depth_uint16(value: object) -> np.ndarray:
depth = np.ascontiguousarray(np.squeeze(to_numpy(value)))
if depth.ndim != 2:
raise ValueError(f"expected depth HW/HW1/1HW, got {depth.shape}")
if depth.dtype == np.int16:
return depth.view(np.uint16)
if depth.dtype == np.uint16:
return depth
if np.issubdtype(depth.dtype, np.floating) and float(np.nanmax(depth)) <= 1.0:
raise TypeError(
"depth was normalized to [0,1]; exact millimetres are unavailable from "
"this loader version. Read the PNG bytes from data/*.parquet instead."
)
return depth.astype(np.uint16, copy=False)
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--repo-id", default="Tachintech/TacRich-Manip")
parser.add_argument("--revision", default="main")
parser.add_argument("--root", type=Path, help="Optional existing local dataset root")
parser.add_argument("--episode-index", type=int, default=0)
parser.add_argument("--frame-index", type=int,
help="Episode-local frame; default is the midpoint")
parser.add_argument("--video-backend", default="pyav")
parser.add_argument("--depth-max-mm", type=float, default=2000.0)
parser.add_argument("--tactile-vmax", type=float,
help="Shared tactile upper limit; default uses current-frame max")
parser.add_argument("--output", type=Path, default=Path("episode_preview.png"))
return parser.parse_args()
def main() -> None:
args = parse_args()
dataset = lerobot_dataset_class()(
repo_id=args.repo_id,
root=args.root,
revision=args.revision,
episodes=[args.episode_index],
video_backend=args.video_backend,
)
frame_index = len(dataset) // 2 if args.frame_index is None else args.frame_index
if not 0 <= frame_index < len(dataset):
raise IndexError(f"frame {frame_index} outside [0, {len(dataset) - 1}]")
if args.depth_max_mm <= 0:
raise ValueError("--depth-max-mm must be positive")
sample = dataset[frame_index]
depth = exact_depth_uint16(sample["observation.depth.cam_front"])
tactile = to_numpy(sample["observation.tactile"]).astype(np.float32)
if tactile.shape[0] != 2:
raise ValueError(f"expected tactile [2,H,W], got {tactile.shape}")
tactile_vmax = args.tactile_vmax
if tactile_vmax is None:
tactile_vmax = max(float(np.nanmax(tactile)), 1.0e-6)
figure, axes = plt.subplots(2, 3, figsize=(16, 9), constrained_layout=True)
for axis, (key, title) in zip(axes[0], RGB_KEYS, strict=True):
axis.imshow(rgb_hwc(sample[key]), interpolation="nearest")
axis.set_title(title)
axis.axis("off")
depth_artist = axes[1, 0].imshow(
depth, cmap="cividis", vmin=0.0, vmax=args.depth_max_mm,
interpolation="nearest"
)
axes[1, 0].set_title("Front depth (lossless uint16)")
axes[1, 0].axis("off")
figure.colorbar(depth_artist, ax=axes[1, 0], label="millimetres", fraction=0.046)
tactile_artists = []
for axis, values, title in zip(
axes[1, 1:], tactile, ("Left tactile", "Right tactile"), strict=True
):
tactile_artists.append(axis.imshow(
values, cmap="magma", vmin=0.0, vmax=tactile_vmax,
interpolation="nearest", aspect="auto"
))
axis.set_title(title)
axis.set_xlabel("sensor column")
axis.set_ylabel("sensor row")
figure.colorbar(
tactile_artists[-1], ax=list(axes[1, 1:]), label="calibrated response",
fraction=0.023
)
state_tip = to_numpy(sample["observation.state_gripper"])
action_tip = to_numpy(sample["action_gripper"])
relative_time = float(to_numpy(sample["timestamp"]).reshape(-1)[0])
figure.suptitle(
f"multiple tasks | episode {args.episode_index} | frame {frame_index} | "
f"t={relative_time:.3f}s\n"
f"tip xyz={np.array2string(state_tip[:3], precision=4)} m | "
f"target xyz={np.array2string(action_tip[:3], precision=4)} m"
)
args.output.parent.mkdir(parents=True, exist_ok=True)
figure.savefig(args.output, dpi=140, facecolor="white")
plt.close(figure)
print(args.output.resolve())
if __name__ == "__main__":
main()
|