Datasets:

Modalities:
Image
Video
ArXiv:
Libraries:
Datasets
License:
vpraveen-nv commited on
Commit
e66e02d
·
verified ·
1 Parent(s): cf80f9d

Update review comments

Browse files
Files changed (1) hide show
  1. README.md +13 -9
README.md CHANGED
@@ -79,7 +79,7 @@ VANTAGE-BENCH/
79
 
80
  Predictions are submitted to the evaluation server hosted on HuggingFace. The server computes metrics against held-out annotations and updates the public leaderboard.
81
 
82
- Evaluation server: TBD
83
 
84
  ## Dataset Format
85
 
@@ -90,7 +90,7 @@ Video (mp4) and Images (jpg).
90
  | Category | Task | Videos | Entries |
91
  |----------|------|--------|---------|
92
  | Semantic | VQA | 296 | 1,257 |
93
- | Semantic | Event Verification | TBD | TBD |
94
  | Temporal | Dense Video Captioning | 104 | 717 |
95
  | Temporal | Temporal Localization | 221 | 1,280 |
96
  | Spatial | 2D Object Localization | 3 | 27,404 bounding boxes (628 frames) |
@@ -102,7 +102,12 @@ Video (mp4) and Images (jpg).
102
  **Total entries (VQA + DVC + Temporal):** 3,254
103
  **Total Data Storage:** 42 GB
104
 
105
- ## References
 
 
 
 
 
106
 
107
  ```bibtex
108
  @inproceedings{Fujita2020SODA,
@@ -127,7 +132,11 @@ Video (mp4) and Images (jpg).
127
  }
128
  ```
129
 
130
- **TBD:** Update with VANTAGE-BENCH paper citation and final HuggingFace repo link.
 
 
 
 
131
 
132
  ## Ethical Considerations
133
 
@@ -137,8 +146,3 @@ Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns [her
137
  ## Changelog
138
 
139
  - **2026-04-14:** Initial dataset release.
140
-
141
- ## Notes & Known Issues
142
-
143
- - Ground truth annotations are not publicly released. All evaluation is performed server-side.
144
- - Some warehouse videos are concatenated clips from longer recording sessions.
 
79
 
80
  Predictions are submitted to the evaluation server hosted on HuggingFace. The server computes metrics against held-out annotations and updates the public leaderboard.
81
 
82
+ Evaluation server: [VANTAGE-Bench](https://github.com/anon-benchmark/VANTAGE-bench)
83
 
84
  ## Dataset Format
85
 
 
90
  | Category | Task | Videos | Entries |
91
  |----------|------|--------|---------|
92
  | Semantic | VQA | 296 | 1,257 |
93
+ | Semantic | Event Verification | 163 | 163 |
94
  | Temporal | Dense Video Captioning | 104 | 717 |
95
  | Temporal | Temporal Localization | 221 | 1,280 |
96
  | Spatial | 2D Object Localization | 3 | 27,404 bounding boxes (628 frames) |
 
102
  **Total entries (VQA + DVC + Temporal):** 3,254
103
  **Total Data Storage:** 42 GB
104
 
105
+ ## Potential Known Risks
106
+
107
+ - Ground truth annotations are not publicly released. All evaluation is performed server-side.
108
+ - Some warehouse videos are concatenated clips from longer recording sessions.
109
+
110
+ ## Citations
111
 
112
  ```bibtex
113
  @inproceedings{Fujita2020SODA,
 
132
  }
133
  ```
134
 
135
+ **TBD:** Update with VANTAGE-BENCH paper citation when published.
136
+
137
+ ## References
138
+
139
+ **TBD:** Add final HuggingFace repo link.
140
 
141
  ## Ethical Considerations
142
 
 
146
  ## Changelog
147
 
148
  - **2026-04-14:** Initial dataset release.