SCIMD-17 (Source Camera Identification — Mobile Devices 17)

📸SCIMD-17 (Source Camera Identification — Mobile Devices 17) is an extended version of SCIMD-6, which originally contained images from 6 mobile devices. This version expands the dataset to 17 smartphone models and ~17,000 real-world images, enabling more comprehensive research in camera model identification, image forensics, and cross-device generalization. Images were collected under varied indoor and outdoor scenes using multiple mobile phone brands to ensure cross-device diversity. The dataset aims to support studies in source camera attribution, cross-device generalization, and mobile imaging behavior.

🧠 Dataset Summary

SCIMD-17 (Smartphone Camera Image Metadata Dataset) is a large-scale dataset containing 17,000 mobile images captured using 17 different smartphone models.This dataset supports research in camera identification, mobile image forensics, image quality analysis, and vision-language model training.

📊 Data Composition

Attribute

Description

Total images

≈17,000

Devices

17 smartphone types

Image type

Real photos captured in natural scenes

Format

JPEG 

Metadata

Make, filename, White Balance, Focal Length, Flash, Model, Date Time Original, Exposure time, ISO Speed Ratings, Exif Offset, Date Time

Use cases

Camera model recognition, forensic analysis, cross-device generalization

🆚 Compared to SCIMD-6

Devices: ↑ from 6 → 17

Images: ↑ from ~6 k → ~17 k

Added new brands and lighting conditions

More balanced per-device sampling

🔬 Research Applications

Smartphone camera model identification

Scene-independent mobile image analysis

Vision-language pre-training (e.g., image captioning or VQA based on EXIF/device clues)

Benchmarking model robustness across devices

🧪 Dataset Collection Methodology

The SCIMD-17 (Source Camera Identification — Mobile Devices 17) dataset was collected to represent real-world smartphone imaging conditions without any laboratory or controlled setup. Images were captured casually in both indoor and outdoor environments, using natural lighting and varied scenes to simulate authentic usage patterns.

Data collection was conducted primarily within the Bapatla Engineering College campus, including classrooms, corridors, and open areas, as well as in residential locations such as houses and surrounding environments. No artificial constraints were imposed on lighting, focus, or scene content, ensuring natural variation typical of everyday photography.

All images were captured manually by volunteers using 17 different smartphone models, listed below:

📱 Smartphone Models Used

📱 Xiaomi_M2101K6P📱 Infinix_Note40_Pro📱 IQZ9X_5G_224📱 MotoG45_5G📱 MotoG64_5G📱 MotoG85_5G📱 Nothing_A001📱 Nothing_Phone1📱 Oneplus_Nord_C3lite📱 Realme_6i📱 Realme8_Pro📱 redmi_9_prime224📱 Redmi14C_5G📱 SamsungM13_5G📱 Vivo_V50_5G📱 VivoT1_5G📱 VivoY56_5G

Captured images were subsequently pre processed to a uniform resolution of 224 × 224 pixels to ensure compatibility with deep learning and vision-language model architectures.

No additional filtering or enhancement was applied beyond resizing, preserving the natural characteristics of each device’s imaging pipeline.

This casual, real-world collection strategy ensures that SCIMD-17 reflects genuine variations in smartphone imaging behavior, making it highly suitable for camera model identification, source attribution, and cross-device generalization studies.

The SCIMD-17 .zip file contains the folders with mobile make as labels. The merged_common17.csv file contains metadata.