Spam Images in Messaging - Annotated Set (SIMAS)

SIMAS Dataset

This archive includes the SIMAS dataset for fine-tuning models for MMS (Multimedia Messaging Service) image moderation. SIMAS is a balanced collection of publicly available images, manually annotated in accordance with a specialized taxonomy designed for identifying visual spam in MMS messages.

Taxonomy for MMS Visual Spam

The following table presents the definitions of categories used for classifying MMS images.

Table 1: Category definitions

Category   
Description 

Alcohol*   
Content related to alcoholic beverages, including advertisements and consumption. 

Drugs*     
Content related to the use, sale, or trafficking of narcotics (e.g., cannabis, cocaine,

Firearms*  
Content involving guns, pistols, knives, or military weapons. 

Gambling*  
Content related to gambling (casinos, poker, roulette, lotteries). 

Sexual     
Content involving nudity, sexual acts, or sexually suggestive material. 

Tobacco*   
Content related to tobacco use and advertisements. 

Violence   
Content showing violent acts, self-harm, or injury. 

Safe       
All other content, including neutral depictions, products, or harmless cultural symbols

Note: Categories marked with an asterisk are regulated in some jurisdictions and may not be universally restricted.

Dataset Collection and Annotation

Data Sources

The SIMAS dataset combines publicly available images from multiple sources, selected to reflect the categories defined in our content taxonomy. Each image was manually reviewed by three independent annotators, with final labels assigned when at least two annotators agreed.

The largest portion of the dataset (30.4%) originates from LAION-400M, a large-scale image-text dataset. To identify relevant content, we first selected a list of ImageNet labels that semantically matched our taxonomy. These labels were generated using GPT-4o in a zero-shot setting, using separate prompts per category. This resulted in 194 candidate labels, of which 88.7% were retained after manual review. The structure of the prompts used in this process is shown in the file gpt4o_imagenet_prompting_scheme.png, which illustrates a shared base prompt template applied across all categories. The fields category_definition, file_examples, and exceptions are specified per category. Definitions align with the taxonomy, while the file_examples column includes sample labels retrieved from the ImageNet label list. The exceptions field contains category-specific filtering instructions; a dash indicates no exceptions were specified.

Another 25.1% of images were sourced from Roboflow, using open datasets such as:

Marijuana and Hemp 200

Drug Detection

Plants Classification

Weapon Detection

Suicide Detection

Violence Detection

Clasificacionimagenes

Waste Recognition

FYP

The NudeNet dataset contributes 11.4% of the dataset. We sampled 1,000 images from the “porn” category to provide visual coverage of explicit sexual content.

Another 11.0% of images were collected from Kaggle, including:

National Flowers

Weapon Dataset for YOLOv5

GUIE Toys

Alcohol Bottle Images

Smoking & Drinking Dataset

An additional 9.9% of images were retrieved from Unsplash, using keyword-based search queries aligned with each category in our taxonomy.

Images from UnsafeBench make up 8.0% of the dataset. Since its original binary labels did not match our taxonomy, all samples were manually reassigned to the most appropriate category.

Finally, 4.2% of images were gathered from various publicly accessible websites. These were primarily used to improve category balance and model generalization, especially in safe classes.

All images collected from the listed sources have been manually reviewed by three independent annotators. Each image is then assigned to a category when at least two annotators reach consensus.

Table 2: Distribution of images per public source and category in SIMAS dataset

Type
Category
LAION
Roboflow
NudeNet
Kaggle
Unsplash
UnsafeBench
Other
Total

Unsafe
Alcohol
29
0
3
267
0
1
0
300

Unsafe
Drugs
17
211
0
0
13
8
1
250

Unsafe
Firearms
0
59
0
229
0
62
0
350

Unsafe
Gambling
132
38
0
0
73
39
18
300

Unsafe
Sexual
2
0
421
0
3
68
6
500

Unsafe
Tobacco
0
446
0
0
43
11
0
500

Unsafe
Violence
0
289
0
0
0
11
0
300

Safe
Alcohol
140
35
0
0
16
13
96
300

Safe
Drugs
67
49
0
15
72
17
30
250

Safe
Firearms
173
15
0
3
144
8
7
350

Safe
Gambling
164
2
0
1
121
12
0
300

Safe
Sexual
235
22
139
2
0
94
8
500

Safe
Tobacco
351
67
5
13
8
16
40
500

Safe
Violence
212
20
3
21
0
42
2
300

All
All
1,522
1,253
571
551
493
402
208
5,000

Balancing

To ensure semantic diversity and dataset balance, undersampling was performed on overrepresented categories using a CLIP-based embedding and k-means clustering strategy. This resulted in a final dataset containing 2,500 spam and 2,500 safe images, evenly distributed across all categories.

Table 3: Distribution of images per category in SIMAS dataset

Type
Alcohol
Drugs
Firearms
Gambling
Sexual
Tobacco
Violence
Total

Unsafe
300
250
350
300
500
500
300
2,500

Safe
300
250
350
300
500
500
300
2,500

All
600
500
700
600
1,000
1,000
600
5,000

SIMAS+ Dataset

For researchers interested in a more realistic deployment setting, we also curate a complementary dataset called SIMAS+. It is a benchmarking dataset containing publicly accessible images extracted from real-world MMS traffic, specifically from external URLs embedded in messages. Manual annotation was conducted by three independent raters, with a category label assigned when at least two annotators agreed. The dataset was then balanced across spam categories using the same semantic grouping strategy as in SIMAS, ensuring equal representation of safe and unsafe examples per class. The final version of SIMAS+ contains 700 images, with the category distribution presented in the table below.

Table 4: Distribution of images per category in SIMAS+ dataset

Type
Alcohol
Drugs
Firearms
Gambling
Sexual
Tobacco
Violence
Total

Unsafe
100
50
80
50
50
10
10
350

Safe
100
50
80
50
50
10
10
350

All
200
100
160
100
100
20
20
700

Note: Due to regulatory and privacy considerations, SIMAS+ is not included in this archive. To obtain access to the SIMAS+ dataset for research purposes, please contact the dataset authors directly.

License
This dataset is licensed under the CC BY-NC 4.0 license and may be used for non-commercial research purposes.

# SIMAS数据集

本归档文件包含用于微调多媒体短信服务（Multimedia Messaging Service, MMS）图像审核模型的SIMAS数据集。SIMAS是一个经过平衡处理的公开图像集合，按照专为识别MMS消息中视觉垃圾信息设计的专业分类体系进行人工标注。

## MMS视觉垃圾信息分类体系
下表列出了用于对MMS图像进行分类的各类别定义。

### 表1：类别定义

| 类别 | 描述 |
| --- | --- |
| 酒类（Alcohol*） | 与酒精饮料相关的内容，包括广告与饮用场景。 |
| 毒品（Drugs*） | 与麻醉药品（如大麻、可卡因）的使用、售卖或贩运相关的内容。 |
| 枪支弹药（Firearms*） | 涉及枪械、手枪、刀具或军用武器的内容。 |
| 赌博（Gambling*） | 与赌博相关的内容（赌场、扑克、轮盘、彩票）。 |
| 色情（Sexual） | 涉及裸露、性行为或性暗示的内容。 |
| 烟草（Tobacco*） | 与烟草使用及广告相关的内容。 |
| 暴力（Violence） | 展示暴力行为、自残或受伤的内容。 |
| 安全（Safe） | 所有其他内容，包括中性描述、产品或无害的文化符号。 |

注：带星号标记的类别在部分司法辖区受监管，未必受到全域性限制。

## 数据集采集与标注
### 数据来源
SIMAS数据集整合了来自多个来源的公开图像，所选图像均匹配我们内容分类体系中的类别。每张图像均由三名独立标注员人工审核，当至少两名标注员达成一致时，即可确定最终标签。

数据集占比最高的部分（30.4%）源自LAION-400M——一个大规模图像-文本数据集。为筛选相关内容，我们首先选取了语义上与我们的分类体系匹配的ImageNet标签。这些标签通过GPT-4o以零样本（Zero-shot）设置生成，每个类别使用独立提示词，最终得到194个候选标签，经人工审核后保留了其中88.7%。本流程中使用的提示词结构详见文件`gpt4o_imagenet_prompting_scheme.png`，该文件展示了适用于所有类别的共享基础提示词模板，其中`category_definition`（类别定义）、`file_examples`（文件示例）和`exceptions`（例外情况）字段均按类别单独指定。定义与本分类体系保持一致，`file_examples`列包含从ImageNet标签列表中提取的示例标签，`exceptions`字段包含特定类别的过滤规则，短横线表示未指定例外情况。

另有25.1%的图像源自Roboflow，使用的公开数据集包括：
- 大麻与工业大麻200（Marijuana and Hemp 200）
- 毒品检测（Drug Detection）
- 植物分类（Plants Classification）
- 武器检测（Weapon Detection）
- 自杀检测（Suicide Detection）
- 暴力检测（Violence Detection）
- 图像分类（Clasificacionimagenes）
- 垃圾识别（Waste Recognition）
- FYP

NudeNet数据集贡献了数据集的11.4%：我们从「porn」类别中采样了1000张图像，以覆盖露骨色情内容的视觉表现。

另有11.0%的图像采集自Kaggle，包括：
- 国花数据集（National Flowers）
- YOLOv5武器数据集（Weapon Dataset for YOLOv5）
- GUIE玩具数据集（GUIE Toys）
- 酒瓶图像数据集（Alcohol Bottle Images）
- 吸烟与饮酒数据集（Smoking & Drinking Dataset）

额外9.9%的图像通过关键词搜索从Unsplash获取，搜索关键词与我们分类体系中的各类别匹配。

UnsafeBench数据集占数据集的8.0%：由于其原始二元标签与我们的分类体系不匹配，所有样本均被人工重新分配至最合适的类别。

最后4.2%的图像采集自各类公开可访问网站，主要用于优化类别平衡性与模型泛化能力，尤其是安全类别。

所有从上述来源采集的图像均由三名独立标注员人工审核，当至少两名标注员达成一致时，即可为每张图像分配类别标签。

### 表2：SIMAS数据集各公开来源与类别的图像分布
| 类型 | 类别 | LAION | Roboflow | NudeNet | Kaggle | Unsplash | UnsafeBench | 其他 | 总计 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 不安全 | 酒类 | 29 | 0 | 3 | 267 | 0 | 1 | 0 | 300 |
| 不安全 | 毒品 | 17 | 211 | 0 | 0 | 13 | 8 | 1 | 250 |
| 不安全 | 枪支弹药 | 0 | 59 | 0 | 229 | 0 | 62 | 0 | 350 |
| 不安全 | 赌博 | 132 | 38 | 0 | 0 | 73 | 39 | 18 | 300 |
| 不安全 | 色情 | 2 | 0 | 421 | 0 | 3 | 68 | 6 | 500 |
| 不安全 | 烟草 | 0 | 446 | 0 | 0 | 43 | 11 | 0 | 500 |
| 不安全 | 暴力 | 0 | 289 | 0 | 0 | 0 | 11 | 0 | 300 |
| 安全 | 酒类 | 140 | 35 | 0 | 0 | 16 | 13 | 96 | 300 |
| 安全 | 毒品 | 67 | 49 | 0 | 15 | 72 | 17 | 30 | 250 |
| 安全 | 枪支弹药 | 173 | 15 | 0 | 3 | 144 | 8 | 7 | 350 |
| 安全 | 赌博 | 164 | 2 | 0 | 1 | 121 | 12 | 0 | 300 |
| 安全 | 色情 | 235 | 22 | 139 | 2 | 0 | 94 | 8 | 500 |
| 安全 | 烟草 | 351 | 67 | 5 | 13 | 8 | 16 | 40 | 500 |
| 安全 | 暴力 | 212 | 20 | 3 | 21 | 0 | 42 | 2 | 300 |
| 总计 | 总计 | 1,522 | 1,253 | 571 | 551 | 493 | 402 | 208 | 5,000 |

### 数据集平衡处理
为确保语义多样性与数据集平衡性，我们针对占比过高的类别采用基于CLIP嵌入与k-means聚类的策略进行欠采样。最终数据集包含2500张垃圾图像与2500张安全图像，所有类别均匀分布。

### 表3：SIMAS数据集各类别图像分布
| 类型 | 酒类 | 毒品 | 枪支弹药 | 赌博 | 色情 | 烟草 | 暴力 | 总计 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 不安全 | 300 | 250 | 350 | 300 | 500 | 500 | 300 | 2,500 |
| 安全 | 300 | 250 | 350 | 300 | 500 | 500 | 300 | 2,500 |
| 总计 | 600 | 500 | 700 | 600 | 1,000 | 1,000 | 600 | 5,000 |

## SIMAS+数据集
针对希望使用更贴近实际部署场景的研究人员，我们还整理了一个互补数据集SIMAS+。该基准数据集包含从真实MMS流量中提取的公开可访问图像，特别是从消息中嵌入的外部URL提取的图像。手动标注由三名独立评分员完成，当至少两名标注员达成一致时分配类别标签。数据集采用与SIMAS相同的语义分组策略实现垃圾信息类别的平衡，确保每个类别中安全与不安全样本的占比均等。最终版本的SIMAS+数据集包含700张图像，类别分布如下表所示。

### 表4：SIMAS+数据集各类别图像分布
| 类型 | 酒类 | 毒品 | 枪支弹药 | 赌博 | 色情 | 烟草 | 暴力 | 总计 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 不安全 | 100 | 50 | 80 | 50 | 50 | 10 | 10 | 350 |
| 安全 | 100 | 50 | 80 | 50 | 50 | 10 | 10 | 350 |
| 总计 | 200 | 100 | 160 | 100 | 100 | 20 | 20 | 700 |

注：由于监管与隐私考量，本归档文件不包含SIMAS+数据集。如需出于研究目的获取SIMAS+数据集，请直接联系数据集作者。

## 许可协议
本数据集采用知识共享署名-非商业性使用4.0国际许可协议（CC BY-NC 4.0）授权，可用于非商业性研究用途。