Research map / Probabilistic and scientific modeling / Variational and generative sampling
FARMER: Flow AutoRegressive Transformer over Pixels
Guangting Zheng, Qinyu Zhao, Tao Yang, Fei Xiao, Zhijie Lin, Jie Wu, Jiajun Deng, Yanyong Zhang, Rui Zhu
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 self-supervised dimension reduction (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | FID: 61.17 -> 49.29 | Improved |
| repeat class embedding 64 times (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | FID: 49.29 -> 45.34 | Improved |
| add token permutation between AF and AR (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | FID: 45.34 -> 44.56 | Improved |
| add naive CFG from JetFormer (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | FID: 44.56 -> 8.66 | Improved |
| use resampling-based CFG instead of naive CFG (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | FID: 8.66 -> 5.67 | Improved |
| resampling CFG lowers IS versus naive CFG (ablation) | FARMER-1.1B · ImageNet 256x256 ablation | IS: 233.84 -> 215.53 | Hurt |
| use autoregressive flow instead of Jet flow (ablation) | FARMER · ImageNet 256x256 NF architecture | FID: 106.23 -> 5.55 | Improved |
| add one-step distillation of AF reverse process | FARMER · ImageNet 256x256 | reverse time per image: 0.1689 -> 0.0076 | Improved |
| add one-step distillation of AF reverse process | FARMER · ImageNet 256x256 | FID: 5.55 -> 5.63 | No clear effect |
| model redundant channels conditioned on informative channels instead of independent Gaussian (JetFormer style) (ablation) | FARMER-1.1B · ImageNet 256x256 | FID: 7.81 -> 5.67 | Improved |
| reduce GMM mixtures to 32 (ablation) | FARMER · ImageNet 256x256 | FID | Hurt |
| increase informative dimension beyond 128 (ablation) | FARMER · ImageNet 256x256 | FID | Hurt |
Closest papers by meaning
- A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation (ICLR 2025 · similarity 0.88)
- STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025 · similarity 0.88)
- Fundamental Limits of Visual Autoregressive Transformers: Universal Approximation Abilities (ICML 2025 · similarity 0.87)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.87)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.87)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.87)
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.
