CCKEB is a benchmark designed for Continual and Compositional Knowledge Editing in Large Vision-Language Models (LVLMs), accepted at NeurIPS 2025.
The benchmark targets realistic knowledge update scenarios in which visual identities and textual facts are edited sequentially.Models are required to retain previously edited knowledge while answering compositional multimodal queries that depend… See the full description on the dataset page: https://huggingface.co/datasets/MemEIC/CCKEB.
