Add a tool that measures emotes from a store's 3D preview

tools/capture drives a store page's skin viewer in headless Firefox, stepping its clock one tick at
a time and taking every tick from ten fixed cameras. It then fits our own rig to the frames by
rendering the model (a Python port of LimbBend and PoseApplier) and matching outlines and colours,
and writes the result as Emotecraft JSON. It measures pixels only and never reads the page's
animation data. The README covers the steps, the checks and the limits.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

tools/capture/synth_test.py

@@ -0,0 +1,85 @@
+"""End-to-end check of measure.py on frames with a known answer: our own twerk test animation, rendered
+by model.py as if it were a capture, then measured like one.
+
+    .venv/bin/python synth_test.py make <capture to copy cameras from> [ticks]   # writes frames/synth
+    .venv/bin/python measure.py synth --ticks 0:4
+    .venv/bin/python synth_test.py check
+"""
+import json
+import sys
+from pathlib import Path
+
+import numpy as np
+from PIL import Image
+from scipy.spatial import cKDTree
+
+from model import INDEX, Model, rest_rig, rig_to_pose
+from render import Camera, render_colors
+
+# Where the fake viewer puts the model: Minecraft's player scale, feet 25 units below the origin.
+CALIB = (0.9375, 0, -1.5, 0)
+SKIN = "Steve"
+OUT = Path("frames/synth")
+
+
+def twerk(t):
+    """TestAnimations.twerk, written in rig terms."""
+    up = 0.5 * (1 - np.cos(t / 8 * 2 * np.pi))
+    thigh, knee, splay, tilt, arch = -1.1 + 0.15 * up, 1.75 - 0.3 * up, 0.55, 1.4, 0.4 + 0.3 * up
+    drop = 12 - 6 * (np.cos(thigh) + np.cos(thigh + knee)) * np.cos(splay)
+    back = -6 * (np.sin(thigh) + np.sin(thigh + knee))
+    chest = tilt - arch
+    v = rest_rig()
+    s = lambda name, value: v.__setitem__(INDEX[name], value)
+    s("hip_y", drop), s("hip_z", back)
+    s("body_pitch", tilt), s("body_bend", arch), s("body_axis", np.pi)
+    s("head_pitch", tilt - chest * 0.5)
+    for arm, roll in (("right_arm", 0.3), ("left_arm", -0.3)):
+        s(f"{arm}_pitch", tilt - chest * 1.1), s(f"{arm}_roll", roll), s(f"{arm}_bend", 0.5)
+    for leg, sign in (("right_leg", 1), ("left_leg", -1)):
+        s(f"{leg}_pitch", thigh), s(f"{leg}_roll", sign * splay), s(f"{leg}_bend", knee)
+        s(f"{leg}_axis", np.pi - sign * 0.35)
+    return v
+
+
+def skin():
+    return np.asarray(Image.open(f"skins/{SKIN}.png").convert("RGBA"))
+
+
+def make(source, ticks):
+    meta = json.loads((Path("frames") / source / "views.json").read_text())
+    model = Model(spacing=0.08, skin=skin())
+    cams = {n: Camera(v, meta["size"], meta["aspect"], 1) for n, v in meta["views"].items()}
+    for tick in range(ticks):
+        pts = np.concatenate(list(model.pose(rig_to_pose(twerk(tick))).values()))
+        for name, (image, mask) in render_colors(pts, model.colors, CALIB, cams).items():
+            (OUT / name).mkdir(parents=True, exist_ok=True)
+            rgba = np.concatenate([image * 255, mask[..., None] * 255], axis=-1).astype(np.uint8)
+            Image.fromarray(rgba).save(OUT / name / f"{tick:04d}.png")
+    meta |= {"url": "synthetic twerk", "skin": SKIN, "ticks": ticks, "loop_ticks": None}
+    (OUT / "views.json").write_text(json.dumps(meta, indent=2))
+    print(f"wrote {ticks} ticks to {OUT}")
+
+
+def check():
+    data = json.loads((OUT / "rig.json").read_text())
+    print("calibration:", np.round(data["calib"], 3), "true:", CALIB)
+    model = Model(skin=skin())
+    for entry in data["ticks"]:
+        truth, got = model.pose(rig_to_pose(twerk(entry["tick"]))), model.pose(rig_to_pose(np.array(entry["rig"])))
+        # The fitted pose sits in the fitted calibration's frame; compare where the viewer shows them.
+        to = lambda p, c: c[0] * p + np.asarray(c[1:]) * [1, -1, -1]
+        parts = []
+        for part in truth:
+            a, b = to(truth[part], CALIB), to(got[part], data["calib"])
+            parts.append((part, np.linalg.norm(a - b, axis=1).mean(), cKDTree(b).query(a)[0].mean()))
+        print(f"tick {entry['tick']}: score {entry['score']:.3f}  "
+              + "  ".join(f"{p} {d:.2f}/{s:.2f}" for p, d, s in parts))
+    print("(per part: mean point-for-point / shape-only distance, in model pixels)")
+
+
+if __name__ == "__main__":
+    if sys.argv[1] == "make":
+        make(sys.argv[2], int(sys.argv[3]) if len(sys.argv) > 3 else 4)
+    else:
+        check()