Research map / Image generation / Text-to-image diffusion
Vector Quantized Diffusion Model for Text-to-Image Synthesis
Check 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 VQ-Diffusion instead of the matched VQ-AR decoder (ablation) | VQ-Diffusion-S on CUB-200 · CUB-200 | FID: 18.12 -> 12.97 | Helped |
| use VQ-Diffusion instead of the matched VQ-AR decoder (ablation) | VQ-Diffusion-B on CUB-200 · CUB-200 | throughput (imgs/s): 0.03 -> 0.13 | Helped |
| reduce inference steps from 100 to 25 (ablation) | VQ-Diffusion-S on CUB-200 · CUB-200 | FID: 12.97 -> 15.46 | Hurt |
| reduce inference steps from 100 to 25 (ablation) | VQ-Diffusion-S on CUB-200 · CUB-200 | throughput (imgs/s): 0.37 -> 1.25 | Helped |
| increase training steps from 10 to 100 (ablation) | VQ-Diffusion on CUB-200 · CUB-200, 100 inference steps | FID: 19.84 -> 11.94 | Helped |
| increase training steps from 100 to 200 (ablation) | VQ-Diffusion on CUB-200 · CUB-200, 100 inference steps | FID: 11.94 -> 12.27 | Hurt |
| reduce inference steps from 200 to 100 (ablation) | VQ-Diffusion on CUB-200 trained for 200 steps · CUB-200 | FID: 11.80 -> 12.27 | No clear effect |
| use mask-and-replace diffusion with final mask rate 0.9 rather than mask-only or replace-only diffusion (ablation) | VQ-Diffusion on Oxford-102 · Oxford-102 | FID | Helped |
| add the auxiliary noiseless-token prediction loss to the variational lower-bound loss (ablation) | VQ-Diffusion · text-to-image synthesis | image quality | Helped |
| use top-token truncation at rate 0.86 during inference (ablation) | VQ-Diffusion on CUB-200 · CUB-200 | FID | Helped |
Closest papers by meaning
- VQ4DiT: Efficient Post-Training Vector Quantization for Diffusion Transformers (AAAI 2025 · similarity 0.89)
- Denoising Autoregressive Transformers for Scalable Text-to-Image Generation (ICLR 2025 · similarity 0.89)
- You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs (ICLR 2025 · similarity 0.89)
- Towards Understanding the Working Mechanism of Text-to-Image Diffusion Model (NeurIPS 2024 · similarity 0.89)
- ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation (ICLR 2025 · similarity 0.88)
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · 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.