JinPLU commited on
Commit
810dfd2
·
verified ·
1 Parent(s): 9e81eda

Simplify human annotations dataset card

Browse files
Files changed (1) hide show
  1. README.md +22 -14
README.md CHANGED
@@ -46,21 +46,29 @@ configs:
46
 
47
  # WRBench Human Annotations
48
 
49
- Human review metadata and labels for WRBench. This release keeps the historical
50
- clean pair boundary and adds the completed 2026-07-07 current-benchmark recheck.
51
 
52
- ## Release Boundary
53
 
54
- - Clean pair metadata rows: 1741 = 1,156 historical rows + 585 current rechecked pairs.
55
- - Distinct `pair_id` values in `pairs`: 1586 (`155` duplicate-id rows are inherited from the historical staging boundary).
56
- - Current recheck submissions: 1755 rows = 585 pairs x 3 annotators.
57
- - Current recheck consensus: 585 rows.
58
- - Exclusions/status rows: 1609 exclusions plus current review status for all 709 pool rows.
59
 
60
- Only current-current, full-six-dimension rows with all p1/p3/p4 submissions and
61
- confirmed current benchmark assets are included as new clean labels. Current-vs-
62
- legacy rows, single-video issue checks, and incomplete rows are kept out of
63
- clean labels and recorded in `exclusions` / `current_review_status`.
 
64
 
65
- Annotator ids are pseudonymized as `annotator_a`, `annotator_b`, etc. The salt is
66
- not included in the public package.
 
 
 
 
 
 
 
 
 
 
 
46
 
47
  # WRBench Human Annotations
48
 
49
+ This dataset contains the human comparison labels used to validate WRBench's
50
+ automatic evaluation metrics.
51
 
52
+ ## Version Update: 2026-07-07
53
 
54
+ We updated the release after rechecking videos that changed during benchmark
55
+ maintenance. The release now includes:
 
 
 
56
 
57
+ - 1,741 clean comparison rows.
58
+ - 4,302 individual human judgments.
59
+ - 585 newly rechecked current-benchmark comparisons, each reviewed by three
60
+ annotators.
61
+ - Majority-label summaries for the newly rechecked subset.
62
 
63
+ ## Configs
64
+
65
+ - `pairs`: clean comparison rows used by the public WRBench release.
66
+ - `annotation_to_video`: maps each comparison to the corresponding video assets.
67
+ - `current_review_consensus`: majority labels for the 585 newly rechecked
68
+ comparisons.
69
+ - `current_review_submissions`: anonymized per-annotator labels for those 585
70
+ comparisons.
71
+ - `current_review_status` and `exclusions`: records kept for transparency, but
72
+ not used as clean comparison labels.
73
+
74
+ Annotator identities are anonymized in the public files.