amazingtrash/scenario-recognition-for-display

---
language:
- en
- zh
license: apache-2.0
tags:
- computer vision
- image classification
- screen scene recognition
- display chip AI
- edge AI
size_categories:
- 10B<n<100B
dataset_info:
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': 3DS
'1': Anime
'2': Apex
'3': CF
'4': CSGO
'5': DeltaForce
'6': Dota2
'7': FJWJ
'8': Face
'9': Forza
'10': Genshin
'11': Kart
'12': LOL
'13': Landscape
'14': Overwatch
'15': PPT
'16': PS
'17': PUBG
'18': QQspeed
'19': Sport
'20': Word
'21': yanyun
splits:
- name: train
pretty_name: Screen Scene Recognition Dataset for Display Chip
---

# Screen Scene Recognition Dataset for Display Chip

## Dataset Description
This dataset is specifically designed for **edge-side AI model development of display chips**, targeting real-time recognition of 22 types of screen scenes. It addresses the pain points of missing public datasets, high category similarity, and poor data quality in screen scene recognition tasks, providing high-quality labeled data for algorithm research and engineering deployment.

### Overview
- **Total Samples**: 53,438 high-quality cleaned images
- **Number of Categories**: 22 distinct screen scenes
- **Average Samples per Category**: ~2,429
- **Image Quality**: All images are processed to remove black borders, duplicate samples (via hash algorithm), and other quality issues

## Dataset Structure
### Categories & Labels
## Dataset Structure
| Category | Subcategory | Quantity (Images) | Dataset Source |
| --- | --- | --- | --- |
| Games | CSGO | 4151 | roboflow: skie-u1yzr/csgo-tr4j7 |
| Games | Delta Force | 2237 | roboflow: yolov11-hs6o5/delta-force-wtowy |
| Games | CrossFire (CF) | 2945 | roboflow: ahmed-kakpz/crossfire-q3x9b<br>roboflow: yoloproject-qkr8c/crossfire-aimbot-pnt3a<br>roboflow: t-kch-enemies-gr/crossfire-enemies-gr<br>roboflow: boyang-atqjy/cf-d1lea<br>roboflow: learning-yn484/cf-tqspc<br>roboflow: cfv10/cf-ktskc<br>roboflow: chris-cbzkl/cf-yrljb<br>roboflow: yiku/cf-yolo11<br>roboflow: lin-swjrf/cf-person<br>roboflow: heavebk/crossfire-zvd56<br>roboflow: cf-twgc7/cf-kfnnp |
| Games | Overwatch | 2291 | roboflow: divertisseur-g7gui/overwatch-sqw1k<br>roboflow: 872858554-qq-com/overwatch-djt7x |
| Games | PUBG | 2533 | roboflow: legendarynuggets-y8wml/pubg-xggpl<br>roboflow: workspace-5ry2i/pubg-imo8q<br>roboflow: gwycc/pubg-ij9vn<br>roboflow: see-ul5qe/pubg-1gopr<br>roboflow: luizconrado/pubg-rhc8l<br>roboflow: 2799283008-qq-com/pubg-jaw28<br>roboflow: projects-r3ul8/pubg-v2ujr<br>roboflow: yolo-hsg3o/pubg-hf77u<br>roboflow: aipubg/pubg-lmzib |
| Games | Apex | 3257 | roboflow: new-workspace-kv0mx/apex-saofm<br>roboflow: 1-0jgxn/apex-s9gtn<br>roboflow: new-workspace-kv0mx/apex-wtj6x<br>roboflow: tristen-2bfd5/apex-ssde6<br>roboflow: edward-ixd04/apex-zbkbp<br>roboflow: auner2456889-gp1gr/apex-rxk8r<br>roboflow: apex-nceuf/apex-x5hna |
| Games | League of Legends (LOL) | 1971 | roboflow: atrashrc-gmail-com/league-of-legends-mtuzl<br>roboflow: glass-cmdst/league-of-legends-wlfmr<br>roboflow: markus-srensen/league-of-legends-7au4o<br>roboflow: league-of-legends-dud09/league-of-legends-ydjt2<br>roboflow: mapleland-sdm9k/league-of-legends-eyiej |
| Games | Dota2 | 2202 | roboflow: sake-rj73p/dota2-qkzjc<br>roboflow: dota-nk9sm/dota2-c1q0w<br>roboflow: laixuxiang/dota2-mwuk0<br>roboflow: fds-exzra/dota2-lasthitter<br>roboflow: d2-a2ij4/augmented-dota2 |
| Games | Fantasy Westward Journey (FWJ) | 1483 | roboflow: yolo-1nxke/-s7y0p<br>roboflow: yolo-1nxke/-eaf1c<br>roboflow: dingsu/-iq8dn |
| Games | Yanyun | 2364 | Screen recording capture |
| Games | Genshin Impact | 3242 | roboflow: will-end-o2a5r/genshin-li3di<br>roboflow: samokat/genshin-uus5a |
| Games | KartRider | 2966 | roboflow: testworkspace/kartrider<br>roboflow: kof98/kartrider-i9xqq<br>Screen recording capture |
| Games | Forza Horizon 4 | 2696 | roboflow: gaurav-tpig5/forza-horizon-4<br>roboflow: robot-my7f9/forza<br>roboflow: placas-n7ft2/forza-gz8bv<br>roboflow: deertracker/forza-roads<br>Screen recording capture |
| Games | QQ Speed | 3650 | Screen recording capture |
| Office | Word | 1952 | huggingface: likaixin/ScreenSpot-Pro<br>Screen recording capture |
| Office | PowerPoint (PPT) | 1046 | Screen recording capture |
| Industrial Software | 3DS | 1430 | Screen recording capture |
| Graphics Software | Photoshop (PS) | 1456 | Screen recording capture |
| Media | Face | 2204 | kaggle: dataturks/face-detection-in-images |
| Media | Landscape | 3237 | kaggle: utkarshsaxenadn/landscape-recognition-image-dataset-12k-images<br>roboflow: master-v2adj/landscape-xtmqp |
| Media | Sport | 2056 | kaggle: sidharkal/sports-image-classification |
| Media | Anime | 2069 | huggingface: yskor/anime_background_city_street<br>huggingface: svjack/Anime_Background_Images<br>huggingface: Sebastian2602/AnimeSceneData<br>roboflow: anime-search/anime-search-g6irh<br>roboflow: anime-search-2/anime-search-2<br>roboflow: oggidetectiondataset/anime-detecter<br>roboflow: hakdog-tmdnj/anime-ebf4j<br>roboflow: juan-pablo-ruiz-flrez/anime-9cqdu<br>roboflow: test-kgq7k/anime-ymci2<br>roboflow: noname-4sril/anime-gun<br>Screen recording capture |

### Data Splits
- **Full Dataset**: 53,438 samples (no predefined train/val/test splits; users are recommended to split according to their own needs)

## Data Collection & Preprocessing
### Data Sources
- **Game Scenes**: Filtered from Roboflow screen target detection datasets (labels removed) and manually captured gameplay footage
- **Office/Productivity**: Manually captured via screen recording (Word, PPT, PS, etc.)
- **Other Scenes**: Collected from public datasets and manual screen captures

### Preprocessing Steps
1. **Black Border Removal**: Cropped invalid black border areas to focus on valid screen content
2. **Deduplication**: Used hash algorithm to eliminate duplicate images
3. **Class Balance**: Applied targeted data augmentation and class weight assignment for imbalanced categories
4. **Quality Control**: Manual cleaning of low-quality/blurry images

## Usage
This dataset is suitable for:
- Research and training of screen scene classification models for edge devices
- Performance comparison of lightweight CNN models (ResNet18, MobileNetV2) on edge AI tasks
- Engineering optimization of display chip-side real-time scene recognition

## License
Apache License 2.0

## Citation
If you use this dataset in your research, please cite:
@dataset {screen_scene_recognition_2026,
author = {amazingtrash},
title = {Screen Scene Recognition Dataset for Display Chip},
year = {2026},
url = {https://huggingface.co/datasets/amazingtrash/scenario-recognition-for-display},
license = {Apache-2.0}}

## Contact
For questions about the dataset, please contact: [e-mail]

language:
- 英语
- 汉语
license: Apache 2.0 许可证
tags:
- 计算机视觉（computer vision）
- 图像分类（image classification）
- 屏幕场景识别（screen scene recognition）
- 显示芯片人工智能（display chip AI）
- 边缘人工智能（edge AI）
size_categories:
- 10B < n < 100B
dataset_info:
features:
- name: 图像（image）
dtype: 图像
- name: 标签（label）
dtype:
类别标签（class_label）:
命名:
'0': 3DS
'1': 动漫（Anime）
'2': Apex
'3': 穿越火线（CF）
'4': 反恐精英：全球攻势（CSGO）
'5': 三角洲部队（DeltaForce）
'6': Dota2
'7': 梦幻西游（Fantasy Westward Journey，简称FWJ，原标注FJWJ）
'8': 人脸（Face）
'9': Forza
'10': 原神（Genshin）
'11': 跑跑卡丁车（Kart）
'12': 英雄联盟（LOL）
'13': 风景（Landscape）
'14': 守望先锋（Overwatch）
'15': 演示文稿（PPT）
'16': Photoshop（PS）
'17': 绝地求生（PUBG）
'18': QQ飞车（QQspeed）
'19': 体育（Sport）
'20': Word
'21': yanyun
splits:
- name: 训练集（train）
pretty_name: 显示芯片屏幕场景识别数据集（Screen Scene Recognition Dataset for Display Chip）
---
# 显示芯片屏幕场景识别数据集（Screen Scene Recognition Dataset for Display Chip）

## 数据集简介
本数据集专为**显示芯片边缘人工智能（edge AI）模型开发**设计，旨在实现22类屏幕场景的实时识别。针对当前屏幕场景识别任务中公开数据集缺失、类别相似度高、数据质量欠佳等痛点，本数据集为算法研究与工程部署提供了高质量的标注数据。

### 概况
- **总样本量**：53438张经过高质量清洗的图像
- **类别数量**：22种独立屏幕场景类别
- **单类别平均样本量**：约2429张
- **图像质量**：所有图像均经过处理，去除黑边、重复样本（通过哈希算法实现）及其他质量问题

## 数据集结构
### 类别与标签
| 大类 | 子类别 | 图像数量 | 数据集来源 |
| --- | --- | --- | --- |
| 游戏（Games） | CSGO | 4151 | Roboflow: skie-u1yzr/csgo-tr4j7 |
| 游戏 | 三角洲部队（Delta Force） | 2237 | Roboflow: yolov11-hs6o5/delta-force-wtowy |
| 游戏 | 穿越火线（CrossFire，简称CF） | 2945 | Roboflow: ahmed-kakpz/crossfire-q3x9b<br>Roboflow: yoloproject-qkr8c/crossfire-aimbot-pnt3a<br>Roboflow: t-kch-enemies-gr/crossfire-enemies-gr<br>Roboflow: boyang-atqjy/cf-d1lea<br>Roboflow: learning-yn484/cf-tqspc<br>Roboflow: cfv10/cf-ktskc<br>Roboflow: chris-cbzkl/cf-yrljb<br>Roboflow: yiku/cf-yolo11<br>Roboflow: lin-swjrf/cf-person<br>Roboflow: heavebk/crossfire-zvd56<br>Roboflow: cf-twgc7/cf-kfnnp |
| 游戏 | 守望先锋（Overwatch） | 2291 | Roboflow: divertisseur-g7gui/overwatch-sqw1k<br>Roboflow: 872858554-qq-com/overwatch-djt7x |
| 游戏 | 绝地求生（PlayerUnknown's Battlegrounds，简称PUBG） | 2533 | Roboflow: legendarynuggets-y8wml/pubg-xggpl<br>Roboflow: workspace-5ry2i/pubg-imo8q<br>Roboflow: gwycc/pubg-ij9vn<br>Roboflow: see-ul5qe/pubg-1gopr<br>Roboflow: luizconrado/pubg-rhc8l<br>Roboflow: 2799283008-qq-com/pubg-jaw28<br>Roboflow: projects-r3ul8/pubg-v2ujr<br>Roboflow: yolo-hsg3o/pubg-hf77u<br>Roboflow: aipubg/pubg-lmzib |
| 游戏 | Apex | 3257 | Roboflow: new-workspace-kv0mx/apex-saofm<br>Roboflow: 1-0jgxn/apex-s9gtn<br>Roboflow: new-workspace-kv0mx/apex-wtj6x<br>Roboflow: tristen-2bfd5/apex-ssde6<br>Roboflow: edward-ixd04/apex-zbkbp<br>Roboflow: auner2456889-gp1gr/apex-rxk8r<br>Roboflow: apex-nceuf/apex-x5hna |
| 游戏 | 英雄联盟（League of Legends，简称LOL） | 1971 | Roboflow: atrashrc-gmail-com/league-of-legends-mtuzl<br>Roboflow: glass-cmdst/league-of-legends-wlfmr<br>Roboflow: markus-srensen/league-of-legends-7au4o<br>Roboflow: league-of-legends-dud09/league-of-legends-ydjt2<br>Roboflow: mapleland-sdm9k/league-of-legends-eyiej |
| 游戏 | Dota2 | 2202 | Roboflow: sake-rj73p/dota2-qkzjc<br>Roboflow: dota-nk9sm/dota2-c1q0w<br>Roboflow: laixuxiang/dota2-mwuk0<br>Roboflow: fds-exzra/dota2-lasthitter<br>Roboflow: d2-a2ij4/augmented-dota2 |
| 游戏 | 梦幻西游（Fantasy Westward Journey，简称FWJ，原标注FJWJ） | 1483 | Roboflow: yolo-1nxke/-s7y0p<br>Roboflow: yolo-1nxke/-eaf1c<br>Roboflow: dingsu/-iq8dn |
| 游戏 | yanyun | 2364 | 屏幕录制捕获 |
| 游戏 | 原神（Genshin Impact，简称Genshin） | 3242 | Roboflow: will-end-o2a5r/genshin-li3di<br>Roboflow: samokat/genshin-uus5a |
| 游戏 | 跑跑卡丁车（KartRider，简称Kart） | 2966 | Roboflow: testworkspace/kartrider<br>Roboflow: kof98/kartrider-i9xqq<br>屏幕录制捕获 |
| 游戏 | 极限竞速（Forza Horizon 4，简称Forza） | 2696 | Roboflow: gaurav-tpig5/forza-horizon-4<br>Roboflow: robot-my7f9/forza<br>Roboflow: placas-n7ft2/forza-gz8bv<br>Roboflow: deertracker/forza-roads<br>屏幕录制捕获 |
| 游戏 | QQ飞车（QQ Speed） | 3650 | 屏幕录制捕获 |
| 办公软件（Office） | Microsoft Word（简称Word） | 1952 | Hugging Face: likaixin/ScreenSpot-Pro<br>屏幕录制捕获 |
| 办公软件 | 演示文稿（PowerPoint，简称PPT） | 1046 | 屏幕录制捕获 |
| 工业软件（Industrial Software） | 3DS | 1430 | 屏幕录制捕获 |
| 图形图像处理软件（Graphics Software） | Photoshop（简称PS） | 1456 | 屏幕录制捕获 |
| 媒体（Media） | 人脸（Face） | 2204 | Kaggle: dataturks/face-detection-in-images |
| 媒体 | 风景（Landscape） | 3237 | Kaggle: utkarshsaxenadn/landscape-recognition-image-dataset-12k-images<br>Roboflow: master-v2adj/landscape-xtmqp |
| 媒体 | 体育（Sport） | 2056 | Kaggle: sidharkal/sports-image-classification |
| 媒体 | 动漫（Anime） | 2069 | Hugging Face: yskor/anime_background_city_street<br>Hugging Face: svjack/Anime_Background_Images<br>Hugging Face: Sebastian2602/AnimeSceneData<br>Roboflow: anime-search/anime-search-g6irh<br>Roboflow: anime-search-2/anime-search-2<br>Roboflow: oggidetectiondataset/anime-detecter<br>Roboflow: hakdog-tmdnj/anime-ebf4j<br>Roboflow: juan-pablo-ruiz-flrez/anime-9cqdu<br>Roboflow: test-kgq7k/anime-ymci2<br>Roboflow: noname-4sril/anime-gun<br>屏幕录制捕获 |

### 数据划分
- **完整数据集**：共53438个样本，未预设训练/验证/测试集划分，建议用户根据自身需求进行划分

## 数据采集与预处理
### 数据来源
- **游戏场景**：从Roboflow屏幕目标检测数据集（已移除原有标签）中筛选，并结合手动捕获的游戏画面
- **办公/生产力场景**：通过屏幕录制手动捕获（Word、PPT、PS等软件界面）
- **其他场景**：从公开数据集及手动屏幕捕获中收集

### 预处理步骤
1. **黑边去除**：裁剪无效黑边区域，聚焦有效屏幕内容
2. **去重处理**：采用哈希算法消除重复图像
3. **类别平衡**：针对不平衡类别应用针对性数据增强及类别权重分配
4. **质量管控**：人工清理低质量、模糊图像

## 应用场景
本数据集适用于：
- 面向边缘设备的屏幕场景分类模型的研究与训练
- 轻量级卷积神经网络（CNN）模型（如ResNet18、MobileNetV2）在边缘人工智能任务上的性能对比
- 显示芯片端实时场景识别算法的工程优化

## 许可证
Apache 许可证 2.0（Apache License 2.0）

## 引用说明
若您在研究中使用本数据集，请引用如下：
bibtex
@dataset{screen_scene_recognition_2026,
author = {amazingtrash},
title = {Screen Scene Recognition Dataset for Display Chip},
year = {2026},
url = {https://huggingface.co/datasets/amazingtrash/scenario-recognition-for-display},
license = {Apache-2.0}
}

## 联系方式
如有数据集相关问题，请联系：[e-mail]