SurFITR: A Dataset for Surveillance Image Forgery Detection and Localisation

SurFITR addresses a highly specific technical gap: the failure of general image forensics to handle surveillance footage (which typically features low resolution, high compression, and small, context-dependent manipulations like removing a person). Its defensibility is currently low (score 4) because it is primarily a research dataset with 0 stars and 3 forks, lacking a community-driven leaderboard or 'data gravity' from massive scale. However, it holds value as a niche reference. Frontier labs (OpenAI, Anthropic) are unlikely to compete directly as this is a forensic/security application rather than a generative or general-reasoning one. The primary threat comes from incumbent cloud providers (AWS Rekognition, Azure Vision) who could integrate similar forensic checks into their security pipelines, or from academic competitors releasing larger, more diverse synthetic datasets. The use of generative models to create the forgeries makes the dataset replicable by anyone with similar compute and prompt engineering expertise.