CrowdBA: A Low-Cost Quality-Driven Crowdsourcing Archi-tecture for Bounding Box Annotation Based on Blockchain

Many blockchain-based crowdsourcing frameworks currently struggle to address the high costs associated with on-chain storage and computation effectively, and they lack a quality-driven in-centive mechanism tailored to bounding box data annotation scenarios. To address these chal-lenges, this paper proposes CrowdBA: a low-cost quality-driven crowdsourcing architecture. The CrowdBA utilizes the Ethereum public blockchain as the foundational architecture and develops corresponding smart contracts. First, by integrating Ethereum with the InterPlanetary File System (IPFS), storage and computation processes are shifted off-chain, effectively addressing the high costs associated with data storage and computation on public blockchains. Additionally, the CrowdBA introduces a Dynamic Intersection over Union Weighted Bounding Box Fusion (DWBF) algorithm, which assigns dynamic weights based on IoU to infer true bounding boxes, thereby assessing each worker's annotation quality. Annotation quality then serves as a key criterion for incentive distribution, ensuring fair and appropriate compensation for all contributors. Experi-mental results demonstrate that the operational costs of each smart contract function remain within reasonable limits; the off-chain storage and computation approach significantly reduces storage and computation expenses, and the DWBF algorithm shows marked improvements in accuracy and robustness over other bounding box fusion methods.