Update review comments
Browse files
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:
|
| 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 |
|
| 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 |
-
##
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|