Datasets:
Tasks:
Image-Text-to-Text
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
Improve dataset card: Update task category, add tags, and add sample usage
#3
by
nielsr
HF Staff
- opened
README.md
CHANGED
@@ -1,4 +1,17 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
dataset_info:
|
3 |
features:
|
4 |
- name: shape
|
@@ -40,8 +53,6 @@ configs:
|
|
40 |
path: data/heptagons_with_visual_cues-*
|
41 |
- split: arrow_on_plus_with_visual_cues
|
42 |
path: data/arrow_on_plus_with_visual_cues-*
|
43 |
-
task_categories:
|
44 |
-
- image-classification
|
45 |
library_name:
|
46 |
- pytorch
|
47 |
---
|
@@ -56,6 +67,19 @@ This dataset is part of the work **"Forgotten Polygons: Multimodal Large Languag
|
|
56 |
|
57 |
This dataset is designed to evaluate the shape understanding capabilities of Multimodal Large Language Models (MLLMs).
|
58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
## Dataset Splits
|
60 |
|
61 |
Each split corresponds to a different reasoning task and shape identification challenge.
|
|
|
1 |
---
|
2 |
+
task_categories:
|
3 |
+
- image-text-to-text
|
4 |
+
tags:
|
5 |
+
- multimodal
|
6 |
+
- mllm
|
7 |
+
- geometric-reasoning
|
8 |
+
- visual-question-answering
|
9 |
+
- shape-recognition
|
10 |
+
- chain-of-thought
|
11 |
+
- mathematics
|
12 |
+
- reasoning
|
13 |
+
language:
|
14 |
+
- en
|
15 |
dataset_info:
|
16 |
features:
|
17 |
- name: shape
|
|
|
53 |
path: data/heptagons_with_visual_cues-*
|
54 |
- split: arrow_on_plus_with_visual_cues
|
55 |
path: data/arrow_on_plus_with_visual_cues-*
|
|
|
|
|
56 |
library_name:
|
57 |
- pytorch
|
58 |
---
|
|
|
67 |
|
68 |
This dataset is designed to evaluate the shape understanding capabilities of Multimodal Large Language Models (MLLMs).
|
69 |
|
70 |
+
## Sample Usage
|
71 |
+
|
72 |
+
This dataset is designed to be used with the evaluation code provided in the [GitHub Repository](https://github.com/rsinghlab/Shape-Blind/tree/main). To evaluate MLLMs on various tasks using this dataset, follow the instructions in the `evaluation` folder of the repository.
|
73 |
+
|
74 |
+
For example, to run a shape identification task using LLaVA-1.5:
|
75 |
+
|
76 |
+
```bash
|
77 |
+
# Navigate to the 'evaluation' folder in the cloned GitHub repository
|
78 |
+
cd Shape-Blind/evaluation
|
79 |
+
# Run the evaluation script
|
80 |
+
python3 evaluate_MLLMs.py --model_version llava-1.5 --task shape_id --dataset_size full
|
81 |
+
```
|
82 |
+
|
83 |
## Dataset Splits
|
84 |
|
85 |
Each split corresponds to a different reasoning task and shape identification challenge.
|