Tracking object states in cooking videos.
Research: project goals and paper reviews. Agents should start with the goals, then read the relevant papers.
- vidosc: VidOSC adaptations and state descriptions.
- tubletegraph: TubeletGraph adaptations and evaluation.
- oscar: OSCAR adaptations and object-state prediction.
- chefOST-data: shared datasets; currently MOSCATO annotations/vocabularies via Git LFS. Shared evaluation tooling and Ego4D data are planned.
- ego4d-pseudo-labels: existing Ego4D pseudo-labeling prototype (private).
VidOSC and TubeletGraph code currently lives on contributor branches, not main.
Use <first-name>/<topic> branches. Keep READMEs concise; link detailed documentation through an index. See the branch and documentation style guide.
VIDOSC-pipeline (private) is deprecated for new work; use vidosc. chefOST-ml (private) retains older mixed experiments; continue model work in the repos above. Raf’s original data pipeline remains on chefOST-data/raf/dataset-generation.
spoc is a research reference; chefOST-site is the demo placeholder (private).
