Research map / Probabilistic and scientific modeling / Molecules and proteins
Sequential Attend, Infer, Repeat: Generative Modelling of Moving Objects
Adam Kosiorek, Hyunjik Kim, Yee Whye Teh, Ingmar Posner
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Add temporal propagation and discovery to AIR | MLP-AIR · Moving multi-MNIST test | log marginal likelihood: 6443.6 -> 6617.6 | Improved |
| Add temporal propagation and discovery to AIR | MLP-AIR · Moving multi-MNIST test | counting accuracy: 0.9058 -> 0.9986 | Improved |
| Add temporal propagation and discovery to AIR | MLP-AIR · Moving multi-MNIST test | addition accuracy: 0.8644 -> 0.9998 | Improved |
| Use convolutional image/glimpse encoders and subpixel-convolution decoder instead of fully connected networks | SQAIR · Moving multi-MNIST test | log marginal likelihood: 6617.6 -> 6784.8 | Improved |
| Use convolutional image/glimpse encoders and subpixel-convolution decoder instead of fully connected networks | SQAIR · Moving multi-MNIST test | addition accuracy: 0.9998 -> 0.9990 | Mixed |
| Remove proposal-glimpse extraction in propagation (ablation) | SQAIR · Moving multi-MNIST | identity consistency | Hurt |
| Use raw observation access in propagation instead of an encoded proposal glimpse (ablation) | SQAIR · Moving multi-MNIST | identity consistency | Hurt |
| Use NVIL instead of VIMCO for discrete-latent gradient estimation | SQAIR · Moving multi-MNIST | training performance | Hurt |
| Learn the output likelihood standard deviation rather than fixing it (ablation) | SQAIR · Moving multi-MNIST | training success | Hurt |
| Use a smaller network with 32 units per layer (ablation) | SQAIR · Moving multi-MNIST | quantitative performance | Hurt |
Closest papers by meaning
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.85)
- Believing is Seeing: Unobserved Object Detection using Generative Models (CVPR 2025 · similarity 0.84)
- SDDGR: Stable Diffusion-based Deep Generative Replay for Class Incremental Object Detection (CVPR 2024 · similarity 0.84)
- ReCon: Region-Controllable Data Augmentation with Rectification and Alignment for Object Detection (NeurIPS 2025 · similarity 0.83)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.83)
- Data-efficient Large Vision Models through Sequential Autoregression (ICML 2024 · similarity 0.83)
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.
