Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -13,4 +13,55 @@ tags:
|
|
13 |
- esk
|
14 |
size_categories:
|
15 |
- 10K<n<100K
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
- esk
|
14 |
size_categories:
|
15 |
- 10K<n<100K
|
16 |
+
task_categories:
|
17 |
+
- question-answering
|
18 |
+
---
|
19 |
+
|
20 |
+
# EPFL Smart Kitchen: Lemonade benchmark
|
21 |
+
|
22 |
+
## Abstract
|
23 |
+
we introduce Lemonade: **L**anguage models **E**valuation of **MO**tion a**N**d **A**ction-**D**riven **E**nquiries.
|
24 |
+
Lemonade consists of 36,521 closed-ended QA pairs linked to egocentric video clips, categorized in three groups and six subcategories.
|
25 |
+
18,857 QAs focus on behavior understanding, leveraging the rich ground truth behavior annotations of the EPFL-Smart Kitchen to interrogate models about perceived actions (Perception) and reason over unseen behaviors (Reasoning).
|
26 |
+
8,210 QAs involve longer video clips, challenging models in summarization (Summarization) and session-level inference (Session properties).
|
27 |
+
The remaining 9,463 QAs leverage the 3D pose estimation data to infer hand shapes, joint angles (Physical attributes), or trajectory velocities (Kinematics) from visual information.
|
28 |
+
|
29 |
+
## Content
|
30 |
+
The current repository contains all egocentric videos recorded in the EPFL-Smart-Kitchen-30 dataset. You can download the rest of the dataset at ... and ... .
|
31 |
+
|
32 |
+
### Repository structure
|
33 |
+
|
34 |
+
```
|
35 |
+
Lemonade
|
36 |
+
βββ MCQs
|
37 |
+
βββ lemonade_benchmark.csv
|
38 |
+
βββ videos
|
39 |
+
βββ YH2002_2023_12_04_10_15_23_hololens.mp4
|
40 |
+
βββ ..
|
41 |
+
βββ README.md
|
42 |
+
```
|
43 |
+
|
44 |
+
`lemonade_benchmark.csv` : Table with the following fields:
|
45 |
+
* **Question** : Question to be answered
|
46 |
+
* **QID** : Question identifier, an integer from 0 to 30
|
47 |
+
* **Answers** : A list of possible answers to the question. This can be a multiple-choice set or open-ended responses.
|
48 |
+
* **Correct Answer** : The answer that is deemed correct from the list of provided answers.
|
49 |
+
* **Clip** : A reference to the video clip related to the question.
|
50 |
+
* **Start** : The timestamp (in frame) in the clip where the question context begins.
|
51 |
+
* **End** : The timestamp (in frame) in the clip where the question context ends.
|
52 |
+
* **Category** : The broad topic under which the question falls (Behavior understanding, Long-term understanding or Motion and Biomechanics)
|
53 |
+
* **Subcategory** : A more refined classification within the category (Perception, Reasoning, Summarization, Session properties, Physical attributes, Kinematics)
|
54 |
+
* **Difficulty** : The complexity level of the question (e.g., Easy, Medium, Hard)
|
55 |
+
|
56 |
+
`videos` : Folder with all egocentric videos from the EPFL-Smart-Kitchen-30 benchmark. Video names are structured as `[Participant_ID]_[Session_name]_hololens.mp4`.
|
57 |
+
|
58 |
+
> See the refered publication for more details on the design of the benchmark.
|
59 |
+
|
60 |
+
## Usage
|
61 |
+
The evaluation of the benchmark can be done through the following github repository: ... .
|
62 |
+
|
63 |
+
## Publications
|
64 |
+
cite arxiv paper
|
65 |
+
|
66 |
+
## Acknowledgments
|
67 |
+
We thank Andy Bonnetto for the design of the dataset and Matea Tashkovska for the adaptation of the evaluation platform.
|