Research map / Image generation / Autoregressive and latent generation
BLT: Bidirectional Layout Transformer for Controllable Layout Generation
Xiang Kong, Lu Jiang, Huiwen Chang, Han Zhang, Yuan Hao, Haifeng Gong, Irfan Essa
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 hierarchical sampling policy (HSP) to non-autoregressive layout decoding (ablation) | BLT · RICO (conditioned on category) | FID: 104 -> 70 | Improved |
| add hierarchical sampling policy (HSP) (ablation) | BLT · RICO (conditioned on category) | IOU: 0.37 -> 0.30 | Improved |
| add hierarchical sampling policy (HSP) (ablation) | BLT · PubLayNet (conditioned on category) | FID: 217 -> 134 | Improved |
| add hierarchical sampling policy (HSP) (ablation) | BLT · PubLayNet (conditioned on category) | Overlap: 0.12 -> 0.04 | Improved |
| generate category first then size then position instead of size then position then category (ablation) | BLT · RICO | IOU: 0.162 -> 0.127 | Improved |
| generate category, then size, then position instead of category, position, size (ablation) | BLT · RICO | IOU: 0.129 -> 0.127 | Improved |
| generate category first rather than size first (ablation) | BLT · RICO | Alignment: 0.351 -> 0.342 | Improved |
| use bidirectional non-autoregressive decoding instead of autoregressive transformer | BLT · RICO (conditioned on category) | FID: 76 -> 70 | Improved |
| use bidirectional decoding instead of autoregressive transformer | BLT · PubLayNet (conditioned on category) | FID: 127 -> 134 | Hurt |
| use bidirectional decoding instead of autoregressive transformer | BLT · COCO (conditioned on category) | IOU: 0.60 -> 0.43 | Improved |
| use BLT instead of autoregressive transformers for unconditional generation | BLT · RICO (unconditional) | IOU: 0.086 -> 0.127 | Hurt |
| use BLT instead of autoregressive transformers for unconditional generation | BLT · COCO (unconditional) | Overlap: 1.709 -> 1.452 | Improved |
Closest papers by meaning
- DLT: Conditioned layout generation with Joint Discrete-Continuous Diffusion Layout Transformer (ICCV 2023 · similarity 0.90)
- MaskPLAN: Masked Generative Layout Planning from Partial Input (CVPR 2024 · similarity 0.88)
- Towards Aligned Layout Generation via Diffusion Model with Aesthetic Constraints (ICLR 2024 · similarity 0.88)
- AesthetiQ: Enhancing Graphic Layout Design via Aesthetic-Aware Preference Alignment of Multi-modal Large Language Models (CVPR 2025 · similarity 0.87)
- Retrieval-Augmented Layout Transformer for Content-Aware Layout Generation (CVPR 2024 · similarity 0.87)
- OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (NeurIPS 2025 · 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.
