Mobile3M: Android Cuttlefish Emulator Images for MobileVLM Pre-training

Mobile3M consists of approximately 1,000 image-based records captured from Android Cuttlefish Emulators for pre-training Mobile Vision Language Models (MobileVLM). Released by Xiaomi Corporation in late 2024, the data supports research into mobile-specific vision-language tasks and UI interaction.

Use Cases

• Pre-training MobileVLM models using Android Cuttlefish screenshots
• Developing UI-based vision-language tasks using the imagefolder format
• Benchmarking mobile-specific pre-training tasks as described in Arxiv 2409.14818

Strengths

• Authored by Xiaomi Corporation researchers
• Standardized imagefolder format for compatibility with the Hugging Face datasets library
• Directly linked to Arxiv 2409.14818 for methodological transparency

Limitations

• Small sample size of under 1,000 records
• Restricted to synthetic Android Cuttlefish Emulator environments which may not reflect physical device nuances
• Non-commercial license (CC BY-NC-SA 4.0) restricts usage

Provenance

Source

Xiaomi Corporation (Arxiv 2409.14818)

Collection Method

Captured from Android Cuttlefish Emulators

Time Range

2024

Freshness

Last updated December 2024; linked to Arxiv publication 2409.14818.