moontidef commited on
Commit
d881dc4
2 Parent(s): da5fcdb 1d0b8be

Merge branch 'main' of https://huggingface.co/datasets/MoonTideF/GridTallyBench

Browse files
Files changed (1) hide show
  1. README.md +124 -125
README.md CHANGED
@@ -1,125 +1,124 @@
1
- ---
2
- language:
3
- - en
4
- license: [mit]
5
- annotations_creators:
6
- - no-annotation
7
- language_creators:
8
- - machine-generated
9
- pretty_name: GridTallyBench
10
- size_categories:
11
- - n<1k
12
- source_datasets:
13
- - original
14
- task_categories:
15
- - image-classification
16
- - object-detection
17
- task_ids:
18
- - multi-class-image-classification
19
- - object-counting
20
- dataset_info:
21
- features:
22
- - name: block_pixel
23
- dtype: int32
24
- - name: grid_size
25
- dtype: int32
26
- - name: first_block
27
- dtype: string
28
- - name: image
29
- dtype: image
30
- splits:
31
- - name: test
32
- num_examples: 960
33
- configs:
34
- - config_name: default
35
- data_files:
36
- - split: test
37
- path: data/test-*
38
- ---
39
-
40
- # GridTallyBench: Checkerboard Image Dataset for MLLM Benchmarking
41
-
42
- ## Overview
43
-
44
- GridTallyBench is a collection of synthetic checkerboard images designed to test and benchmark Multi-modal Large Language Models (MLLMs) on tasks involving visual pattern recognition and counting. This dataset offers a controlled environment for evaluating model performance on basic visual tasks, particularly useful for assessing an MLLM's ability to count and describe simple geometric patterns.
45
-
46
- ## Dataset Details
47
-
48
- - **Name**: GridTallyBench
49
- - **Version**: 1.0.0
50
- - **Task**: Image classification and object counting
51
- - **Size**: 960 images
52
- - **Format**: Parquet file containing image data and metadata
53
- - **License**: MIT
54
-
55
- ## Content
56
-
57
- The dataset consists of checkerboard images with the following variations:
58
-
59
- - **Block sizes**: 1x1 to 24x24 pixels
60
- - **Grid sizes**: 1x1 to 20x20 blocks
61
- - **Starting colors**: Black-first and white-first patterns
62
-
63
- Each image in the dataset is accompanied by metadata including:
64
-
65
- - `block_pixel`: Size of each square in pixels (1 to 24)
66
- - `grid_size`: Number of squares in each row/column (1 to 20)
67
- - `first_block`: Color of the top-left square ('black' or 'white')
68
- - `image`: Binary data of the PNG image
69
-
70
- ## Use Cases
71
-
72
- This dataset is particularly useful for:
73
-
74
- 1. Testing MLLM's ability to count objects in images
75
- 2. Evaluating pattern recognition capabilities
76
- 3. Assessing color differentiation in simple scenarios
77
- 4. Benchmarking performance on controlled, synthetic images
78
-
79
- ## Loading the Dataset
80
-
81
- To load and use this dataset with the Hugging Face `datasets` library:
82
-
83
- ```python
84
- from datasets import load_dataset
85
-
86
- dataset = load_dataset("MoonTideF/GridTallyBench")
87
-
88
- # Access the first item
89
- first_item = dataset['test'][0]
90
- print(f"Block size: {first_item['block_pixel']}x{first_item['block_pixel']} pixels")
91
- print(f"Grid size: {first_item['grid_size']}x{first_item['grid_size']} blocks")
92
- print(f"First block color: {first_item['first_block']}")
93
- dataset['test'][0]['image'].show()
94
- ```
95
-
96
- ## Dataset Creation
97
-
98
- This dataset was generated using a custom Python script. The images are synthetic and do not contain any real-world content or personal information.
99
-
100
- ## Limitations
101
-
102
- - The dataset is limited to black and white colors only
103
- - Images are synthetic and may not represent real-world complexity
104
- - The largest image size is 480x480 pixels (20x20 grid with 24x24 pixel blocks)
105
-
106
- ## Citation
107
-
108
- If you use this dataset in your research, please cite it as follows:
109
-
110
- ```
111
- @misc{gridtallybench,
112
- author = {MoonTideF},
113
- title = {GridTallyBench: Checkerboard Image Dataset for MLLM Benchmarking},
114
- year = {2024},
115
- publisher = {Hugging Face},
116
- journal = {Hugging Face Datasets},
117
- howpublished = {\url{https://huggingface.co/datasets/MoonTideF/GridTallyBench}}
118
- }
119
- ```
120
-
121
- ## Contact
122
-
123
- For any questions or feedback regarding this dataset, please contact [Your Contact Information].
124
-
125
- ---
 
1
+ ---
2
+ language:
3
+ - en
4
+ license: [mit]
5
+ annotations_creators:
6
+ - no-annotation
7
+ language_creators:
8
+ - machine-generated
9
+ pretty_name: GridTallyBench
10
+ size_categories:
11
+ - n<1k
12
+ source_datasets:
13
+ - original
14
+ task_categories:
15
+ - image-classification
16
+ - object-detection
17
+ task_ids:
18
+ - multi-class-image-classification
19
+ dataset_info:
20
+ features:
21
+ - name: block_pixel
22
+ dtype: int32
23
+ - name: grid_size
24
+ dtype: int32
25
+ - name: first_block
26
+ dtype: string
27
+ - name: image
28
+ dtype: image
29
+ splits:
30
+ - name: test
31
+ num_examples: 960
32
+ configs:
33
+ - config_name: default
34
+ data_files:
35
+ - split: test
36
+ path: data/test-*
37
+ ---
38
+
39
+ # GridTallyBench: Checkerboard Image Dataset for MLLM Benchmarking
40
+
41
+ ## Overview
42
+
43
+ GridTallyBench is a collection of synthetic checkerboard images designed to test and benchmark Multi-modal Large Language Models (MLLMs) on tasks involving visual pattern recognition and counting. This dataset offers a controlled environment for evaluating model performance on basic visual tasks, particularly useful for assessing an MLLM's ability to count and describe simple geometric patterns.
44
+
45
+ ## Dataset Details
46
+
47
+ - **Name**: GridTallyBench
48
+ - **Version**: 1.0.0
49
+ - **Task**: Image classification and object counting
50
+ - **Size**: 960 images
51
+ - **Format**: Parquet file containing image data and metadata
52
+ - **License**: MIT
53
+
54
+ ## Content
55
+
56
+ The dataset consists of checkerboard images with the following variations:
57
+
58
+ - **Block sizes**: 1x1 to 24x24 pixels
59
+ - **Grid sizes**: 1x1 to 20x20 blocks
60
+ - **Starting colors**: Black-first and white-first patterns
61
+
62
+ Each image in the dataset is accompanied by metadata including:
63
+
64
+ - `block_pixel`: Size of each square in pixels (1 to 24)
65
+ - `grid_size`: Number of squares in each row/column (1 to 20)
66
+ - `first_block`: Color of the top-left square ('black' or 'white')
67
+ - `image`: Binary data of the PNG image
68
+
69
+ ## Use Cases
70
+
71
+ This dataset is particularly useful for:
72
+
73
+ 1. Testing MLLM's ability to count objects in images
74
+ 2. Evaluating pattern recognition capabilities
75
+ 3. Assessing color differentiation in simple scenarios
76
+ 4. Benchmarking performance on controlled, synthetic images
77
+
78
+ ## Loading the Dataset
79
+
80
+ To load and use this dataset with the Hugging Face `datasets` library:
81
+
82
+ ```python
83
+ from datasets import load_dataset
84
+
85
+ dataset = load_dataset("MoonTideF/GridTallyBench")
86
+
87
+ # Access the first item
88
+ first_item = dataset['test'][0]
89
+ print(f"Block size: {first_item['block_pixel']}x{first_item['block_pixel']} pixels")
90
+ print(f"Grid size: {first_item['grid_size']}x{first_item['grid_size']} blocks")
91
+ print(f"First block color: {first_item['first_block']}")
92
+ dataset['test'][0]['image'].show()
93
+ ```
94
+
95
+ ## Dataset Creation
96
+
97
+ This dataset was generated using a custom Python script. The images are synthetic and do not contain any real-world content or personal information.
98
+
99
+ ## Limitations
100
+
101
+ - The dataset is limited to black and white colors only
102
+ - Images are synthetic and may not represent real-world complexity
103
+ - The largest image size is 480x480 pixels (20x20 grid with 24x24 pixel blocks)
104
+
105
+ ## Citation
106
+
107
+ If you use this dataset in your research, please cite it as follows:
108
+
109
+ ```
110
+ @misc{gridtallybench,
111
+ author = {MoonTideF},
112
+ title = {GridTallyBench: Checkerboard Image Dataset for MLLM Benchmarking},
113
+ year = {2024},
114
+ publisher = {Hugging Face},
115
+ journal = {Hugging Face Datasets},
116
+ howpublished = {\url{https://huggingface.co/datasets/MoonTideF/GridTallyBench}}
117
+ }
118
+ ```
119
+
120
+ ## Contact
121
+
122
+ For any questions or feedback regarding this dataset, please contact [Your Contact Information].
123
+
124
+ ---