Add task category to metadata
#2
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,5 +1,7 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
|
| 5 |
# Omni-DuplexEval
|
|
@@ -10,8 +12,8 @@ Omni-DuplexEval is a benchmark for evaluating real-time duplex multimodal intera
|
|
| 10 |
|
| 11 |
The benchmark contains two scenarios:
|
| 12 |
|
| 13 |
-
* Real-Time Description (RTD): evaluates continuous streaming description ability.
|
| 14 |
-
* Proactive Reminder (PR): evaluates event-aware and proactive interaction ability.
|
| 15 |
|
| 16 |
Omni-DuplexEval contains 660 samples in total, including 300 samples for Real-Time Description and 360 samples for Proactive Reminder. The dataset consists of 9 tasks with human-curated timestamp-level annotations.
|
| 17 |
|
|
@@ -59,20 +61,22 @@ Each sample contains the following fields:
|
|
| 59 |
|
| 60 |
### Annotation Details
|
| 61 |
|
| 62 |
-
Question Audio
|
| 63 |
|
| 64 |
-
question_audio is aligned with the full video duration.
|
| 65 |
* For RTD tasks, the question is asked at the beginning of the video.
|
| 66 |
* For PR tasks, the question may occur at arbitrary timestamps.
|
| 67 |
-
Reference Answers
|
| 68 |
-
* answer1 and answer2 provide human-annotated reference responses.
|
| 69 |
-
* In correction tasks, answer1 additionally contains the corrected target response.
|
| 70 |
-
* In PR_event_reminder and PR_post_event_reminder, the answer fields are empty.
|
| 71 |
|
| 72 |
-
|
| 73 |
|
| 74 |
-
*
|
| 75 |
-
* In correction tasks,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
* In RTD tasks, reminder fields are empty.
|
| 77 |
|
| 78 |
## Data Collection and Safety
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- video-text-to-text
|
| 5 |
---
|
| 6 |
|
| 7 |
# Omni-DuplexEval
|
|
|
|
| 12 |
|
| 13 |
The benchmark contains two scenarios:
|
| 14 |
|
| 15 |
+
* **Real-Time Description (RTD)**: evaluates continuous streaming description ability.
|
| 16 |
+
* **Proactive Reminder (PR)**: evaluates event-aware and proactive interaction ability.
|
| 17 |
|
| 18 |
Omni-DuplexEval contains 660 samples in total, including 300 samples for Real-Time Description and 360 samples for Proactive Reminder. The dataset consists of 9 tasks with human-curated timestamp-level annotations.
|
| 19 |
|
|
|
|
| 61 |
|
| 62 |
### Annotation Details
|
| 63 |
|
| 64 |
+
**Question Audio**
|
| 65 |
|
| 66 |
+
`question_audio` is aligned with the full video duration.
|
| 67 |
* For RTD tasks, the question is asked at the beginning of the video.
|
| 68 |
* For PR tasks, the question may occur at arbitrary timestamps.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
+
**Reference Answers**
|
| 71 |
|
| 72 |
+
* `answer1` and `answer2` provide human-annotated reference responses.
|
| 73 |
+
* In correction tasks, `answer1` additionally contains the corrected target response.
|
| 74 |
+
* In `PR_event_reminder` and `PR_post_event_reminder`, the answer fields are empty.
|
| 75 |
+
|
| 76 |
+
**Reminder Timestamps**
|
| 77 |
+
|
| 78 |
+
* `reminder1` and `reminder2` store timestamp annotations related to reminder events.
|
| 79 |
+
* In correction tasks, `reminder1` contains the timestamp of the incorrect user statement/event.
|
| 80 |
* In RTD tasks, reminder fields are empty.
|
| 81 |
|
| 82 |
## Data Collection and Safety
|