TheKernel01 commited on
Commit
3a84681
·
verified ·
1 Parent(s): ac112ed

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +83 -0
README.md CHANGED
@@ -39,3 +39,86 @@ configs:
39
  - split: validation
40
  path: data/validation-*
41
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39
  - split: validation
40
  path: data/validation-*
41
  ---
42
+
43
+ # Tiny GenImage Dataset
44
+
45
+ ## 📝 Dataset Description
46
+
47
+ ### Dataset Summary
48
+
49
+ The Tiny GenImage Dataset is a curated, scaled-down collection of images and associated metadata designed to train, validate, and benchmark models for detecting and identifying artificially generated content. The dataset contains a mix of real-world images alongside those generated by prominent AI models, including various diffusion models (like Stable Diffusion 1.4/1.5, GLIDE, Midjourney, ADM, VQDM, Wukong) and GANs (BigGAN).
50
+
51
+ Each image is labeled under two categories, enabling researchers and developers to tackle two distinct, high-value computer vision tasks: binary real/fake classification and multi-class source model identification.
52
+
53
+ ### Supported Tasks and Leaderboards
54
+
55
+ This dataset directly supports two critical image classification tasks:
56
+
57
+ |**Task ID**|**Task Name**|**Description**|**Output Classes**|
58
+ |---|---|---|---|
59
+ |**Task A**|Binary Veracity Classification|Classifying images as either real or fake.|2 (real, fake)|
60
+ |**Task B**|AI Model Source Identification|Identifying the specific AI generation model used for images labeled as AI-Generated.|9 (Real, ADM, BigGAN, GLIDE, Midjourney, SD14, SD15, VQDM, Wukong)|
61
+
62
+ ### Languages
63
+
64
+ The descriptive text, including all class labels and metadata, is in English (en).
65
+
66
+ ## 🗂️ Data Splits
67
+
68
+ The dataset is divided into training and validation splits to facilitate standard machine learning workflows.
69
+
70
+ |**Split**|**Number of Instances**|**Notes**|
71
+ |---|---|---|
72
+ |**train**|28,000|Used for model training and weight optimization.|
73
+ |**validation**|7,000|Used for hyperparameter tuning and intermediate model evaluation.|
74
+
75
+ ## 💾 Dataset Structure
76
+
77
+ ### Data Instances
78
+
79
+ A single data instance consists of an image file and two distinct labels detailing its source and authenticity.
80
+
81
+ |**Field Name**|**Example Value**|**Description**|
82
+ |---|---|---|
83
+ |**image**|`<PIL.Image.Image object>`|The actual image content loaded into a PIL object.|
84
+ |**label**|`1`|Binary label for authenticity (Real vs. AI-Generated).|
85
+ |**generator**|`4`|Multi-class label for the specific generation model (or Real).|
86
+
87
+ ### Data Fields
88
+
89
+ The dataset contains the following fields:
90
+
91
+ |**Field Name**|**Data Type**|**Description**|
92
+ |---|---|---|
93
+ |**image**|`datasets.Image()`|The actual image content (e.g., .jpg, .png).|
94
+ |**label**|`datasets.ClassLabel`|Task A: Binary label for image veracity.|
95
+ |**generator**|`datasets.ClassLabel`|Task B: Label specifying the generation source/model.|
96
+
97
+ ## 🏷️ Label Definitions
98
+
99
+ The two label fields use the following strict mappings:
100
+
101
+ **`label` (Binary Veracity Classification)**
102
+
103
+ |**Label**|**Value**|**Description**|
104
+ |---|---|---|
105
+ |**real**|`0`|Image is a real photograph/non-AI generated.|
106
+ |**fake**|`1`|Image was created by an AI generation model.|
107
+
108
+ **`generator` (Model Source Identification)**
109
+
110
+ |**Label**|**Value**|**Description**|
111
+ |---|---|---|
112
+ |**Real**|`0`|Real image (no AI generation involved).|
113
+ |**ADM**|`1`|Generated by Ablated Diffusion Model (Guided Diffusion).|
114
+ |**BigGAN**|`2`|Generated by BigGAN.|
115
+ |**GLIDE**|`3`|Generated by GLIDE.|
116
+ |**Midjourney**|`4`|Generated by Midjourney.|
117
+ |**SD14**|`5`|Generated by Stable Diffusion 1.4.|
118
+ |**SD15**|`6`|Generated by Stable Diffusion 1.5.|
119
+ |**VQDM**|`7`|Generated by Vector Quantized Diffusion Model.|
120
+ |**Wukong**|`8`|Generated by the Wukong diffusion model.|
121
+
122
+ ## 🔗 Sources
123
+
124
+ - **Original dataset:** [yangsangtai/tiny-genimage (Kaggle)](https://www.kaggle.com/datasets/yangsangtai/tiny-genimage)