Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,74 @@
|
|
1 |
-
---
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- computer-vision
|
4 |
+
- visual-reasoning
|
5 |
+
- puzzle
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
license: cc-by-4.0
|
9 |
+
pretty_name: "CLEVR-Sudoku"
|
10 |
+
dataset_info:
|
11 |
+
task_categories:
|
12 |
+
- visual-reasoning
|
13 |
+
task_ids:
|
14 |
+
- puzzle-solving
|
15 |
+
---
|
16 |
+
|
17 |
+
# CLEVR-Sudoku
|
18 |
+
|
19 |
+
## Dataset Summary
|
20 |
+
CLEVR-Sudoku is a **challenging visual puzzle dataset** requiring both visual object perception and reasoning capabilities. Each sample contains:
|
21 |
+
- A partially filled Sudoku puzzle presented with **CLEVR-based** imagery.
|
22 |
+
- Separate “options” images that illustrate how specific **object properties** map to digits.
|
23 |
+
- Metadata specifying puzzle attributes and the ground-truth solution.
|
24 |
+
|
25 |
+
Designed to encourage **visual reasoning** and **pattern recognition**, this dataset provides 6 different subsets (no further train/val/test splits). Each subset contains 1,000 puzzles, leading to **6,000** total puzzles.
|
26 |
+
|
27 |
+
---
|
28 |
+
|
29 |
+
## Supported Tasks
|
30 |
+
- **Visual Reasoning**: Models must interpret compositional 3D object scenes to solve Sudoku logic.
|
31 |
+
- **Pattern Recognition**: Identify how object attributes (shape, color, size, etc.) correlate with digit placements.
|
32 |
+
|
33 |
+
---
|
34 |
+
|
35 |
+
## Dataset Structure
|
36 |
+
|
37 |
+
### Data Instances
|
38 |
+
|
39 |
+
A typical data instance has the following structure:
|
40 |
+
|
41 |
+
```python
|
42 |
+
{
|
43 |
+
"sudoku": 9x9 array of images or None,
|
44 |
+
"options": 9x5 or 9x10 array of images,
|
45 |
+
"attributes": dict that maps each digit to an attribute combination,
|
46 |
+
"id": integer identifier for the puzzle,
|
47 |
+
"solution": 9x9 array of integers (full sudoku)
|
48 |
+
}
|
49 |
+
```
|
50 |
+
|
51 |
+
## Data Splits / Subsets
|
52 |
+
|
53 |
+
There are 6 different subsets, 3 for CLEVR-Easy (CLEVR-Easy-K10, CLEVR-Easy-K30, CLEVR-Easy-K50) and 3 for CLEVR (CLEVR-4-K10, CLEVR-4-K30, CLEVR-4-K50). For CLEVR-Easy only color and shape are relevant for the digit mapping while for CLEVR also material and size relevant are. K indicates how many cells are empty in the sudoku, i.e. K10 means that there are 10 empty cells. Each subset contains 1,000 puzzles. Currently, there are no separate training/validation/test splits within each subset.
|
54 |
+
|
55 |
+
# Dataset Creation
|
56 |
+
|
57 |
+
## Curation Rationale
|
58 |
+
|
59 |
+
Goal: Combine the compositional “CLEVR”-style visual complexity with the logical constraints of Sudoku. This setup pushes models to perform both visual recognition (understanding shapes, colors, etc.) and abstract reasoning (solving Sudoku).
|
60 |
+
|
61 |
+
## Source Data
|
62 |
+
|
63 |
+
Synthetic Generation: The images are created in a CLEVR-like manner, programmatically generated with variations in shape, color, position, etc.
|
64 |
+
Sudoku Logic: Each puzzle is automatically generated and there exists exactly one solution to the puzzle.
|
65 |
+
|
66 |
+
## Annotations
|
67 |
+
|
68 |
+
Automatic Generation: Since the dataset is synthetic, the puzzle solutions are known programmatically. No human annotation is required for the puzzle solutions.
|
69 |
+
Attributes: Each digit (1–9) is associated with one or more visual properties (e.g., color = "red", shape = "cube"). These are also generated systematically.
|
70 |
+
|
71 |
+
## Personal and Sensitive Information
|
72 |
+
|
73 |
+
None: The dataset is purely synthetic, containing no personal or demographic data.
|
74 |
+
|