EgoMonth is a month-level egocentric video question-answering benchmark for evaluating long-term spatiotemporal memory in multimodal large language models. The dataset focuses on daily-life first-person videos and QA tasks that require temporal indexing, spatial grounding, multi-video reasoning, and long-horizon memory.
This repository provides QA metadata, structured annotations, representative anonymized sample videos, and baseline… See the full description on the dataset page: https://huggingface.co/datasets/anonymous-egomonth/egomonth-dataset.
