Add benchmark test subset and clip-length note

README.md

@@ -42,8 +42,10 @@ articulation, temporal dynamics, and spatial relations.
 | Captions | 5 per sequence: tag, short, medium, detailed, paragraph |
 | License | CC BY-NC 4.0, non-commercial research use only |
 
-The benchmark in the paper evaluates on the 17,348 test clips of 40–196 frames, the length
-range used for training.
+Clips range from 30 to 600 frames. The paper's benchmark scores only the 17,348 test clips of
+40–196 frames (listed in `splits/benchmark_test.txt`); evaluating on the full test split will not
+reproduce the paper's numbers. During training, clips longer than 196 frames are randomly cropped
+to 196 frames and clips shorter than 40 frames are zero-padded to 40.
 
 ## Files
 
@@ -78,6 +80,7 @@ split (15,152 videos are shared by train and test). For leakage-free use, `split
 - `video_level/{train,val,test}.txt`: a strict per-source-video split (461,203 / 53,935 /
   26,791 segments) for training new models; each video is assigned by
   `sha256(video_id) mod 100` (<85 train, <95 val, else test).
+- `benchmark_test.txt`: the 17,348 test segments of 40–196 frames used by the paper's benchmark.
 - `split_stats.json`: segment and video counts, and the video-id rule.
 
 Segment ids match each sample's `sample_path`; the source video id is the prefix before

splits/split_stats.json

@@ -38,5 +38,9 @@
       "segments": 26791,
       "videos": 12228
     }
+  },
+  "benchmark_test": {
+    "description": "Released test segments of 40-196 frames, scored by the paper's benchmark.",
+    "segments": 17348
   }
 }