Research map / Video / Video segmentation and tracking
DynaProto: Dynamic Prototypical Contrast for Temporally Consistent Object-Centric Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes DynaProto, a dynamic prototypical contrast method for unsupervised video object-centric learning that enforces temporal consistency beyond adjacent frames.
Problem
Unsupervised video object-centric learning needs temporally consistent object representations, but existing methods assume adjacent frames share the same object composition, which fails for real videos where objects persist over long spans while composition changes.
Approach
Replaces the local adjacent-frame consistency assumption with dynamic prototypical contrast that tracks object prototypes across longer temporal spans, allowing object composition to change over time.
Closest papers by meaning
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.91)
- Cycle Consistency Driven Object Discovery (ICLR 2024 · similarity 0.88)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.87)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.87)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.87)
- Parallelized Spatiotemporal Slot Binding for Videos (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.