From 8f5ebe2aeb4a7f10fb5bf1bbffa7b0d5b5b644df Mon Sep 17 00:00:00 2001
From: shaoxiongduan <shaoxiongduan@gmail.com>
Date: Thu, 7 May 2026 06:14:55 +0000
Subject: [PATCH] =?UTF-8?q?[bugfix]=20eval:=20human=5Faction=20=E2=80=94?=
 =?UTF-8?q?=20load=20kinetics=20labels=20from=20upstream=20submodule?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The metric resolved the Kinetics-400 label file via a wm-eval-era
``_third_party/umt/kinetics_400_categories.txt`` path that no longer
exists after the vbench-as-submodule port. ``os.path.exists`` was
silently False, leaving the label dict empty; every prediction then
mapped to the empty string and every score returned 0.0.

Resolve the path through ``vbench.third_party.umt.__file__`` instead,
which always points at the pinned upstream submodule.
---
 .../metrics/vbench/human_action/metric.py     | 32 +++++++++----------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/fastvideo/eval/metrics/vbench/human_action/metric.py b/fastvideo/eval/metrics/vbench/human_action/metric.py
index 2879a04f4..6bb10f70b 100644
--- a/fastvideo/eval/metrics/vbench/human_action/metric.py
+++ b/fastvideo/eval/metrics/vbench/human_action/metric.py
@@ -7,7 +7,7 @@ Score = 1.0 if match found, 0.0 otherwise.
 
 from __future__ import annotations
 
-import os
+from pathlib import Path
 
 import torch
 from torchvision.transforms.functional import resize, center_crop, normalize
@@ -18,26 +18,26 @@ from fastvideo.eval.types import MetricResult
 from fastvideo.eval.io.video import extract_frames
 
 
-# Kinetics-400 class names (loaded lazily)
-_CAT_DICT = None
+# Kinetics-400 class names (loaded lazily). The label file ships inside
+# the upstream vbench submodule.
+_CAT_DICT: dict[str, str] | None = None
 
 
-def _load_cat_dict():
+def _load_cat_dict() -> dict[str, str]:
     global _CAT_DICT
     if _CAT_DICT is not None:
         return _CAT_DICT
-    cat_path = os.path.join(
-        os.path.dirname(__file__), "..", "_third_party", "umt",
-        "kinetics_400_categories.txt",
-    )
-    _CAT_DICT = {}
-    if os.path.exists(cat_path):
-        with open(cat_path) as f:
-            for line in f:
-                parts = line.strip().split("\t")
-                if len(parts) == 2:
-                    cat, idx = parts
-                    _CAT_DICT[idx] = cat.lower()
+    import vbench.third_party.umt as _umt_pkg
+    cat_path = (Path(_umt_pkg.__file__).resolve().parent
+                / "kinetics_400_categories.txt")
+    out: dict[str, str] = {}
+    with cat_path.open() as f:
+        for line in f:
+            parts = line.strip().split("\t")
+            if len(parts) == 2:
+                cat, idx = parts
+                out[idx] = cat.lower()
+    _CAT_DICT = out
     return _CAT_DICT