BiliSakura/RSCCM
RSCCM is a 7 billion parameter vision-language model developed by BiliSakura, fine-tuned from Qwen2.5-VL-7B-Instruct with a 32768 token context length. It specializes in remote sensing change captioning, specifically designed to describe disaster events from bi-temporal satellite imagery. This model excels at generating news-style descriptions of changes observed between pre- and post-event remote sensing images.
Loading preview...
RSCCM: Remote Sensing Change Captioning Model
RSCCM is a 7 billion parameter vision-language model developed by BiliSakura, specifically fine-tuned for remote sensing change captioning. Built upon the Qwen2.5-VL-7B-Instruct architecture, this model leverages the large-scale RSCC dataset to identify and describe changes in satellite imagery, particularly those related to disaster events.
Key Capabilities
- Specialized Change Captioning: Generates detailed, news-style descriptions of changes between two satellite images (pre- and post-event).
- Disaster Event Analysis: Optimized for understanding and articulating the impact of disaster events as observed in remote sensing data.
- High Context Length: Inherits a 32768 token context length, allowing for comprehensive analysis of visual and textual inputs.
- Supervised Full-Tuning: Achieves its specialized performance through supervised full-tuning on a dedicated dataset.
Good For
- Disaster Response and Monitoring: Automatically generating textual summaries of damage or changes in affected areas.
- Geospatial Intelligence: Extracting actionable insights from bi-temporal remote sensing imagery.
- Environmental Monitoring: Identifying and describing environmental shifts over time.
- Research in Remote Sensing: A foundational model for further research in remote sensing image analysis and captioning.