Research map / Detection and segmentation / Open-set and few-shot detection
Simple Unsupervised Object-Centric Learning for Complex and Naturalistic Videos
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use transformer-based reconstruction instead of mixture-based decoding | STEVE versus SAVi and OP3 · Textured video datasets | Image FG-ARI | Helped |
| Train the slot-transformer model on videos rather than single frames | STEVE versus SLATE · All evaluated datasets | Image FG-ARI | Helped |
| Combine temporal learning with the slot-transformer decoder | STEVE versus SLATE · MOVi-Tex | Image FG-ARI | Helped |
| Use STEVE for video segmentation instead of the baselines | STEVE · Traffic and Aquarium | Video FG-ARI | Helped |
| Use STEVE rather than the baselines | STEVE · CATER | Image FG-ARI | Mixed |
| Use larger dVAE patches instead of the default 4×4 patches (ablation) | STEVE · Evaluated datasets | Video FG-ARI | Hurt |
| Replace the Spatial Broadcast decoder with a deconvolution-based CNN decoder (ablation) | SAVi · Five textured datasets | Video FG-ARI | No clear effect |
| Use decoding masks instead of input attention masks (ablation) | SAVi · Evaluated datasets | Video FG-ARI | Helped |
| Increase video length at test time from 6 to 24 (ablation) | STEVE · CATERTex, MOVi-D, and MOVi-E | Video FG-ARI | Hurt |
| Test on a higher number of objects than in training (ablation) | STEVE · Out-of-distribution test sets | Video FG-ARI | No clear effect |
| Test on novel textures (ablation) | STEVE · MOVi-Tex | Video FG-ARI | Hurt |
| Introduce camera motion (ablation) | STEVE · MOVi-E versus MOVi-D | Video FG-ARI | Helped |
Closest papers by meaning
- Object-Centric Learning for Real-World Videos by Predicting Temporal Feature Similarities (NeurIPS 2023 · similarity 0.90)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.89)
- Provably Learning Object-Centric Representations (ICML 2023 · similarity 0.88)
- VONet: Unsupervised Video Object Learning With Parallel U-Net Attention and Object-wise Sequential VAE (ICLR 2024 · similarity 0.87)
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.86)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.85)
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.