Datasets:
Formats:
parquet
Languages:
English
Size:
< 1K
ArXiv:
Tags:
multi-modal-qa
geometry-qa
abstract-reasoning
geometry-reasoning
visual-puzzle
non-verbal-reasoning
License:
Kexuan Sun
commited on
Commit
•
88ea1b1
1
Parent(s):
49f9bb4
Update README.md and answer labels
Browse files- README.md +19 -1
- marvel_illustration.jpeg +3 -0
- marvel_labels.jsonl +0 -0
README.md
CHANGED
@@ -1,5 +1,7 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
3 |
task_categories:
|
4 |
- visual-question-answering
|
5 |
language:
|
@@ -12,6 +14,7 @@ size_categories:
|
|
12 |
### Dataset Description
|
13 |
|
14 |
MARVEL is a new comprehensive benchmark dataset that evaluates multi-modal large language models' abstract reasoning abilities in six patterns across five different task configurations, revealing significant performance gaps between humans and SoTA MLLMs.
|
|
|
15 |
|
16 |
### Dataset Sources [optional]
|
17 |
|
@@ -27,6 +30,21 @@ Evaluations for multi-modal large language models' abstract reasoning abilities.
|
|
27 |
|
28 |
The directory **images** keeps all images, and the file **marvel_labels.jsonl** provides annotations and explanations for all questions.
|
29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
## Citation [optional]
|
31 |
|
32 |
**BibTeX:**
|
@@ -37,4 +55,4 @@ The directory **images** keeps all images, and the file **marvel_labels.jsonl**
|
|
37 |
journal={arXiv preprint arXiv:2404.13591},
|
38 |
year={2024}
|
39 |
}
|
40 |
-
```
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
paperswithcode_id: marvel
|
4 |
+
pretty_name: MARVEL (Multidimensional Abstraction and Reasoning through Visual Evaluation and Learning)
|
5 |
task_categories:
|
6 |
- visual-question-answering
|
7 |
language:
|
|
|
14 |
### Dataset Description
|
15 |
|
16 |
MARVEL is a new comprehensive benchmark dataset that evaluates multi-modal large language models' abstract reasoning abilities in six patterns across five different task configurations, revealing significant performance gaps between humans and SoTA MLLMs.
|
17 |
+
![image](./marvel_illustration.jpeg)
|
18 |
|
19 |
### Dataset Sources [optional]
|
20 |
|
|
|
30 |
|
31 |
The directory **images** keeps all images, and the file **marvel_labels.jsonl** provides annotations and explanations for all questions.
|
32 |
|
33 |
+
### Fields
|
34 |
+
|
35 |
+
- **id** is of ID of the question
|
36 |
+
- **pattern** is the high-level pattern category of the question
|
37 |
+
- **task_configuration** is the task configuration of the question
|
38 |
+
- **avr_question** is the text of the AVR question
|
39 |
+
- **answer** is the answer to the AVR question
|
40 |
+
- **explanation** is the textual reasoning process to answer the question
|
41 |
+
- **f_perception_question** is the fine-grained perception question
|
42 |
+
- **f_perception_answer** is the answer to the fine-grained perception question
|
43 |
+
- **f_perception_distractor** is the distractor of the fine-grained perception question
|
44 |
+
- **c_perception_question_tuple** is a list of coarse-grained perception questions
|
45 |
+
- **c_perception_answer_tuple** is a list of answers to the coarse-grained perception questions
|
46 |
+
- **file** is the path to the image of the question
|
47 |
+
|
48 |
## Citation [optional]
|
49 |
|
50 |
**BibTeX:**
|
|
|
55 |
journal={arXiv preprint arXiv:2404.13591},
|
56 |
year={2024}
|
57 |
}
|
58 |
+
```
|
marvel_illustration.jpeg
ADDED
Git LFS Details
|
marvel_labels.jsonl
CHANGED
The diff for this file is too large to render.
See raw diff
|
|