Upload folder using huggingface_hub

app.py

@@ -89,6 +89,12 @@ def _cleanup_stale_tmp() -> None:
 
 _cleanup_stale_tmp()
 
+_REPO_ROOT = Path(__file__).resolve().parents[2]
+PROCESSED_DAVIS_DIR = Path(os.environ.get(
+    "PROCESSED_DAVIS_DIR",
+    str(_REPO_ROOT / "data" / "processed" / "DAVIS"),
+))
+
 DAVIS_PALETTE = np.array([
     [  0,   0,   0], [128,   0,   0], [  0, 128,   0], [128, 128,   0],
     [  0,   0, 128], [128,   0, 128], [  0, 128, 128], [128, 128, 128],
@@ -102,6 +108,33 @@ DEFAULT_ALPHA = 0.55
 DEFAULT_CRF   = 18
 THUMB_W, THUMB_H = 427, 240   # 16:9 thumbnails (half of 854×480 DAVIS frames)
 
+# ── VC experiment subset ───────────────────────────────────────────────────────
+# Canonical 10-pair evaluation set defined in exp_016/config_davis.yaml and
+# reused verbatim in exp_018 and exp_019.
+# Taxonomy axes: context (similar/different) · category · movement
+VC_SUBSET: list[tuple[str, list[tuple[str, str]]]] = [
+    ("Class 1 — similar context · similar category · similar movement", [
+        ("blackswan",          "mallard-water"),
+    ]),
+    ("Class 2 — similar context · similar category · different movement", [
+        ("paragliding-launch", "paragliding"),
+        ("motocross-bumps",    "motocross-jump"),
+        ("mallard-fly",        "mallard-water"),
+    ]),
+    ("Class 5 — different context · similar category · similar movement", [
+        ("car-roundabout",     "bus"),
+        ("car-turn",           "car-shadow"),
+        ("lucia",              "hike"),
+        ("breakdance-flare",   "breakdance"),
+    ]),
+    ("Class 6 — different context · similar category · different movement", [
+        ("longboard",          "kite-surf"),
+    ]),
+    ("Class 8 — different context · different category · different movement", [
+        ("blackswan",          "boat"),
+    ]),
+]
+
 # ── Dataset download ───────────────────────────────────────────────────────────
 
 HF_CACHE_REPO = "emirkisa/DAVIS-2017-480p-mp4"    # pre-encoded MP4s
@@ -838,7 +871,51 @@ def build_ui():
 """)
 
             # ──────────────────────────────────────────────────────────────
-            # Tab 5 · About
+            # Tab 5 · VC Subset
+            # ──────────────────────────────────────────────────────────────
+            with gr.TabItem("🔬 VC Subset"):
+                gr.Markdown(
+                    "### DAVIS VC Experiment Subset — 10 Pairs\n"
+                    "Canonical evaluation set used across **exp_016 / exp_018 / exp_019**.  \n"
+                    "Each clip is `data/processed/DAVIS/<seq>/2s.mp4` — 48 frames @ 24 fps.  \n\n"
+                    "**Class taxonomy** (context · category · movement):\n\n"
+                    "| Class | Context | Category | Movement | # Pairs |\n"
+                    "|-------|---------|----------|----------|---------|\n"
+                    "| 1 | similar | similar | similar | 1 |\n"
+                    "| 2 | similar | similar | different | 3 |\n"
+                    "| 3 | similar | different | similar | 0 |\n"
+                    "| 4 | similar | different | different | 0 |\n"
+                    "| 5 | different | similar | similar | 4 |\n"
+                    "| 6 | different | similar | different | 1 |\n"
+                    "| 7 | different | different | similar | 0 |\n"
+                    "| 8 | different | different | different | 1 |"
+                )
+
+                def _vc_label(seq: str) -> str:
+                    if seq not in ALL_SEQUENCES:
+                        return seq
+                    r = DF[DF["sequence"] == seq].iloc[0]
+                    return f"{seq}  [{r['frames']}f · {r['n_objects']}obj]"
+
+                for _cls_label, _pairs in VC_SUBSET:
+                    gr.Markdown(f"---\n#### {_cls_label}")
+                    for _start_seq, _end_seq in _pairs:
+                        _s_path = PROCESSED_DAVIS_DIR / _start_seq / "2s.mp4"
+                        _e_path = PROCESSED_DAVIS_DIR / _end_seq   / "2s.mp4"
+                        with gr.Row():
+                            gr.Video(
+                                value=str(_s_path) if _s_path.exists() else None,
+                                label=f"start · {_vc_label(_start_seq)}",
+                                autoplay=True, loop=True, height=280,
+                            )
+                            gr.Video(
+                                value=str(_e_path) if _e_path.exists() else None,
+                                label=f"end · {_vc_label(_end_seq)}",
+                                autoplay=True, loop=True, height=280,
+                            )
+
+            # ──────────────────────────────────────────────────────────────
+            # Tab 6 · About
             # ──────────────────────────────────────────────────────────────
             with gr.TabItem("ℹ️ About"):
                 gr.Markdown(f"""