Research map / Image generation / Autoregressive and latent generation
Binary Latent Diffusion
Ze Wang, Jiang Wang, Zicheng Liu, Qiang Qiu
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 |
|---|---|---|---|
| Use binary latent diffusion with 64 steps instead of the strongest reported baseline | Image generation · LSUN Churches 256×256 | FID: 3.86 -> 4.36 | Hurt |
| Use binary latent diffusion with 64 steps instead of the strongest reported baseline | Image generation · FFHQ 256×256 | FID: 4.16 -> 5.85 | Hurt |
| Use binary latent diffusion with 64 steps instead of the strongest reported baseline | Image generation · LSUN Bedrooms 256×256 | FID: 2.35 -> 3.85 | Hurt |
| Use binary latent diffusion with 64 steps instead of the strongest reported baseline | Image generation · CelebA-HQ 256×256 | FID: 5.11 -> 6.24 | Hurt |
| Use binary latent diffusion instead of absorbing diffusion to model binary latents | Binary latent sampler · LSUN Churches 256×256 | FID: 5.44 -> 4.36 | Improved |
| Use binary latent diffusion instead of absorbing diffusion to model binary latents | Binary latent sampler · FFHQ 256×256 | FID: 7.64 -> 5.85 | Improved |
| Represent image patches with binary latents rather than VQ codes | Image autoencoder · LSUN Bedrooms validation, 256×256, 16×16 latent grid | PSNR: 20.08 ± 1.84 -> 24.02 ± 2.11 | Improved |
| Represent image patches with binary latents rather than VQ codes | Image autoencoder · LSUN Bedrooms validation, 256×256, 16×16 latent grid | SSIM: 0.62 ± 0.09 -> 0.75 ± 0.06 | Improved |
| Use binary latent diffusion sampling instead of LDM | Image generation · Sampling speed | seconds per sample: 15.68 -> 0.20 | Improved |
| Set λ to 0.1 instead of 0 in the combined objective (ablation) | Binary Latent Diffusion · LSUN Churches 256×256 | FID: 5.07 -> 4.36 | Improved |
| Set λ to 1.0 instead of 0.1 in the combined objective (ablation) | Binary Latent Diffusion · LSUN Churches 256×256 | FID: 4.36 -> 5.12 | Hurt |
| Predict the binary residual zᵗ ⊕ z⁰ rather than z⁰ (ablation) | Binary Latent Diffusion · LSUN Churches 256×256 | FID: 4.72 -> 4.36 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- MaskGIT — MaskGIT: Masked Generative Image Transformer (CVPR 2022) · better on 0 of 1 shared result
- StyleSwin — StyleSwin: Transformer-Based GAN for High-Resolution Image Generation (CVPR 2022) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.90)
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.89)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.89)
- Isometric Representation Learning for Disentangled Latent Space of Diffusion Models (ICML 2024 · similarity 0.89)
- Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models (NeurIPS 2025 · similarity 0.89)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.88)
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.
