Add dataset card, copyright notice, indexes and tooling

scripts/build_dataset.py

@@ -0,0 +1,120 @@
+#!/usr/bin/env python3
+"""Build the dataset indexes: walk every .glb once and write both CSVs.
+
+Reads only each file's glTF JSON chunk, so the whole collection is probed in
+seconds regardless of total size.
+
+    python3 build_dataset.py [--src DIR] [--link]
+
+--link additionally hardlinks the source assets into ../glb/.
+"""
+import argparse, csv, json, os, re, shutil, sys, time
+HERE = os.path.dirname(os.path.abspath(__file__)); sys.path.insert(0, HERE)
+from glb_probe import read_gltf_json, probe
+
+ROOT = os.path.dirname(HERE)
+XL   = re.compile(r"^([0-9a-f]{24})_(fbx|glb|gltf)\.glb$")
+SKFB = re.compile(r"^([0-9a-f]{32})\.glb$")
+
+ASSET_COLS = ["file", "object_id", "id_family", "source_format", "size_bytes",
+              "num_vertices", "num_meshes", "num_nodes", "num_skins", "num_joints",
+              "num_skeleton_roots", "single_tree", "num_animations",
+              "total_duration_sec", "max_keyframes", "animated_joints", "generator"]
+CLIP_COLS  = ["file", "object_id", "clip_index", "clip_name", "duration_sec",
+              "keyframes", "num_channels", "animated_nodes", "animated_joints",
+              "drives_skeleton"]
+
+
+def identify(fname):
+    m = XL.match(fname)
+    if m:
+        return m.group(1), m.group(2), "objaverse_xl"
+    s = SKFB.match(fname)
+    if s:
+        return s.group(1), "", "objaverse_sketchfab"
+    return os.path.splitext(fname)[0], "", "unknown"
+
+
+def clips_of(gltf, joints):
+    accs = gltf.get("accessors", []) or []
+    for i, a in enumerate(gltf.get("animations", []) or []):
+        dur = kf = 0
+        for smp in (a.get("samplers") or []):
+            k = smp.get("input")
+            if k is None or k >= len(accs):
+                continue
+            acc = accs[k]
+            kf = max(kf, acc.get("count", 0) or 0)
+            mx = acc.get("max")
+            if isinstance(mx, list) and mx:
+                try: dur = max(dur, float(mx[0]))
+                except (TypeError, ValueError): pass
+        tgt = {(c.get("target") or {}).get("node") for c in (a.get("channels") or [])}
+        tgt.discard(None)
+        yield {"clip_index": i, "clip_name": (a.get("name") or "")[:120],
+               "duration_sec": round(dur, 4), "keyframes": kf,
+               "num_channels": len(a.get("channels") or []),
+               "animated_nodes": len(tgt), "animated_joints": len(tgt & joints)}
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--src", default=os.path.join(ROOT, "glb"))
+    ap.add_argument("--link", action="store_true", help="hardlink --src assets into ../glb/")
+    args = ap.parse_args()
+
+    if args.link and os.path.abspath(args.src) != os.path.abspath(os.path.join(ROOT, "glb")):
+        dst = os.path.join(ROOT, "glb"); os.makedirs(dst, exist_ok=True)
+        for f in sorted(os.listdir(args.src)):
+            if f.endswith(".glb") and not os.path.exists(os.path.join(dst, f)):
+                try: os.link(os.path.join(args.src, f), os.path.join(dst, f))
+                except OSError: shutil.copy2(os.path.join(args.src, f), os.path.join(dst, f))
+        args.src = dst
+
+    files = sorted(f for f in os.listdir(args.src) if f.endswith(".glb"))
+    t0 = time.time(); assets = []; clips = []; failed = []
+    for n, f in enumerate(files, 1):
+        path = os.path.join(args.src, f)
+        oid, fmt, fam = identify(f)
+        try:
+            info = probe(path)
+            _, gltf = read_gltf_json(path)
+        except Exception as e:
+            failed.append((f, str(e)[:80])); continue
+        joints = set()
+        for s in (gltf.get("skins", []) or []):
+            joints.update(s.get("joints") or [])
+        cs = list(clips_of(gltf, joints))
+        for c in cs:
+            c.update(file="glb/" + f, object_id=oid,
+                     drives_skeleton=str(c["animated_joints"] > 0).lower())
+            clips.append(c)
+        assets.append({
+            "file": "glb/" + f, "object_id": oid, "id_family": fam, "source_format": fmt,
+            "size_bytes": os.path.getsize(path),
+            "num_vertices": info["num_vertices"], "num_meshes": info["num_meshes"],
+            "num_nodes": info["num_nodes"], "num_skins": info["num_skins"],
+            "num_joints": info["num_joints"], "num_skeleton_roots": info["num_skeleton_roots"],
+            "single_tree": str(info["num_skeleton_roots"] == 1).lower(),
+            "num_animations": info["num_animations"],
+            "total_duration_sec": round(sum(c["duration_sec"] for c in cs), 4),
+            "max_keyframes": info["max_keyframes"], "animated_joints": info["animated_joints"],
+            "generator": info["generator"],
+        })
+        if n % 2000 == 0:
+            print("  %d/%d  %.0fs" % (n, len(files), time.time() - t0), flush=True)
+
+    for name, cols, rows in (("metadata.csv", ASSET_COLS, assets),
+                             ("animations.csv", CLIP_COLS, clips)):
+        with open(os.path.join(ROOT, name), "w", newline="", encoding="utf-8") as fh:
+            w = csv.DictWriter(fh, fieldnames=cols); w.writeheader()
+            w.writerows({k: r.get(k, "") for k in cols} for r in rows)
+        print("%-15s -> %d rows" % (name, len(rows)))
+    if failed:
+        print("failed: %d" % len(failed))
+        for f, e in failed[:5]: print("  ", f, e)
+    print("%.0fs" % (time.time() - t0))
+
+
+if __name__ == "__main__":
+    main()