Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -41,3 +41,55 @@ configs:
|
|
41 |
- split: arrow_on_plus_with_visual_cues
|
42 |
path: data/arrow_on_plus_with_visual_cues-*
|
43 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
- split: arrow_on_plus_with_visual_cues
|
42 |
path: data/arrow_on_plus_with_visual_cues-*
|
43 |
---
|
44 |
+
# Forgotten Polygons: Multimodal Large Language Models are Shape-Blind
|
45 |
+
|
46 |
+
This dataset is part of the work **"Forgotten Polygons: Multimodal Large Language Models are Shape-Blind"**.
|
47 |
+
📖 **[Read the Paper](https://www.arxiv.org/abs/2502.15969)**
|
48 |
+
💾 **[GitHub Repository](https://github.com/rsinghlab/Shape-Blind/tree/main)**
|
49 |
+
|
50 |
+
## Overview
|
51 |
+
|
52 |
+
This dataset is designed to evaluate the shape understanding capabilities of Multimodal Large Language Models (MLLMs).
|
53 |
+
|
54 |
+
## Dataset Splits
|
55 |
+
|
56 |
+
Each split corresponds to a different reasoning task and shape identification challenge.
|
57 |
+
|
58 |
+
### 🟢 **Regular Polygons (`regular_polygons`)**
|
59 |
+
- Task: **Shape Identification & Sides Counting**
|
60 |
+
- Description: Consists of images of **regular polygons** (e.g., triangles, pentagons, hexagons, etc.).
|
61 |
+
- Example Queries:
|
62 |
+
- *"What shape is in the image?"*
|
63 |
+
- *"How many sides does the shape in the image have?"*
|
64 |
+
|
65 |
+
### 🟡 **Regular Polygon Pairs (`regular_polygon_pairs`)**
|
66 |
+
- Task: **Multi-Shape Reasoning**
|
67 |
+
- Description: Images contain **two distinct polygons**. The task involves **identifying both shapes, counting their sides, and summing the total**.
|
68 |
+
- Example Query:
|
69 |
+
- *"What are the two shapes in the image, and how many sides do they have in total?"*
|
70 |
+
|
71 |
+
### 🔵 **Abstract Shapes (`abstract_shapes`)**
|
72 |
+
- Task: **Complex Shape Recognition**
|
73 |
+
- Description: Features **irregular and merged polygons**, stars, arrows, and abstract geometric figures.
|
74 |
+
- Example Query:
|
75 |
+
- *"How many sides does this shape have?"*
|
76 |
+
|
77 |
+
### 🟣 **Heptagons with Visual Cues (`heptagons_with_visual_cues`)**
|
78 |
+
- Task: **Visually-Cued Chain-of-Thought (VC-CoT) Reasoning**
|
79 |
+
- Description: Evaluates **VC-CoT prompting** by annotating it with **visual cues** on top of heptagon images.
|
80 |
+
- We chose heptagons because it was the most difficult regular polygon for MLLMs.
|
81 |
+
- The annotations range from ordered numbers and letters, to random numbers and letters.
|
82 |
+
- Example Query:
|
83 |
+
- *"Observe the shape and list the numbers you see. How many sides does the shape have?"*
|
84 |
+
|
85 |
+
### 🔴 **Arrow on Plus with Visual Cues (`arrow_on_plus_with_visual_cues`)**
|
86 |
+
- Task: **VC-CoT with Alternative Visual Cues**
|
87 |
+
- Description: Similar to the **heptagons_with_visual_cues** split but using **arrow-on-plus shapes** instead.
|
88 |
+
- Example Query:
|
89 |
+
- *"Count the total number of numbers associated with the shape’s sides."*
|
90 |
+
|
91 |
+
## Citation
|
92 |
+
If you use this dataset, please cite:
|
93 |
+
|
94 |
+
> Forgotten Polygons: Multimodal Large Language Models are Shape-Blind
|
95 |
+
> [Arxiv: 2502.15969](https://www.arxiv.org/abs/2502.15969)
|