File size: 3,050 Bytes
8420318 | 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 | """End-to-end demo: download one episode, run every toolbox utility, and save
a montage (raw | diff heatmap | contact overlay | depth) plus a quick console
report. Doubles as the end-to-end verification.
python -m react_toolbox.demo # default motherboard episode
python -m react_toolbox.demo --task pushT --episode 2026-06-18/episode_000
"""
from __future__ import annotations
import argparse
import numpy as np
import react_toolbox as T
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--repo", default="yxma/React")
ap.add_argument("--task", default="motherboard")
ap.add_argument("--episode", default="2026-05-10/episode_000")
ap.add_argument("--side", default="left", choices=["left", "right"])
ap.add_argument("--n", type=int, default=200)
ap.add_argument("--out", default="toolbox_demo.png")
ap.add_argument("--with_depth", action="store_true")
args = ap.parse_args()
from huggingface_hub import hf_hub_download
date, ep = args.episode.split("/")
vid = hf_hub_download(args.repo, f"data/{args.task}/videos/{date}/{ep}/tactile_{args.side}.mp4", repo_type="dataset")
pqf = hf_hub_download(args.repo, f"data/{args.task}/meta/{date}/{ep}.parquet", repo_type="dataset")
meta = T.load_meta(pqf)
inten = meta[f"tactile_{args.side}_intensity"][:args.n]
frames = T.load_video(vid, range(args.n))
ref = T.get_reference(frames, mode="p01", intensity=inten)
i = int(np.argmax(inten)) # peak-contact frame
frame = frames[i]
# signals
metrics = T.contact_metrics(frame, ref)
mask = T.contact_mask(frame, ref)
centroid = T.contact_centroid(mask)
print(f"[demo] {args.task} {args.episode} side={args.side} frame={i}")
print(f" contact metrics: {metrics}")
print(f" contact mask area={mask.mean()*100:.1f}% centroid={centroid}")
panels = [frame,
T.diff_heatmap(frame, ref),
T.contact_overlay(frame, ref),
T.reference_compare(frame, ref)[:, :640]] # ref panel for size match
labels = ["raw tactile", "diff heatmap", "contact overlay", "reference"]
if args.with_depth:
from react_toolbox import depth, viz
h = depth.height_map(frame, ref)
panels.append(viz.depth_view(h)); labels.append("depth (approx)")
print(f" depth: range=[{h.min():.2f},{h.max():.2f}] in-contact={h[mask].mean():.2f} out={h[~mask].mean():.2f}")
# actions
pose = meta[f"sensor_{args.side}_pose"][:args.n]
act = T.next_state_action(pose); dlt = T.delta_pose_action(pose)
print(f" actions: next_state {act.shape}, delta {dlt.shape}")
try:
import cv2
montage = np.concatenate([cv2.resize(p, (320, 240)) for p in panels], axis=1)
cv2.imwrite(args.out, montage[..., ::-1]) # RGB->BGR for cv2
print(f" saved montage -> {args.out} ({', '.join(labels)})")
except Exception as e:
print(f" (montage skipped: {e})")
print("[demo] OK")
if __name__ == "__main__":
main()
|