Research map / Image generation / Autoregressive and latent generation
CART: Compositional AutoRegressive Transformer for Image Generation
Siddharth Roheda, Rohit Chowdhury, Aniruddha Bala, Rohan Jaiswal
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 base-detail token decomposition (next-detail) replacing next-scale | VAR-d30 · ImageNet 256x256 | FID: 1.70 -> 1.57 | Improved |
| add base-detail token decomposition at small scale | VAR-d16 · ImageNet 256x256 | FID: 3.30 -> 2.89 | Improved |
| use multi-scale tokenization for base factor (ablation) | CART-d16 · ImageNet 256x256 | FID: 2.90 -> 2.89 | No clear effect |
| increase decomposition order from 0 to 3 (ablation) | CART-d30 · ImageNet 256x256 | FID: 1.70 -> 1.57 | Improved |
| increase decomposition order from 3 to 4 (ablation) | CART-d30 · ImageNet 256x256 | FID: 1.57 -> 1.60 | Hurt |
| add rejection sampling | CART-d30 · ImageNet 256x256 | FID: 1.65 -> 1.57 | Improved |
| train at 256 and generate 512 with patchwise detail prediction (overlapping) | CART-256 · ImageNet 512x512 | FID: 2.63 (VAR-d36-s trained at 512) -> 2.54 | Improved |
| use non-overlapping instead of 50% overlapping patches (ablation) | CART-256 · ImageNet 512x512 | FID: 2.54 -> 2.85 | Hurt |
| use CART for super-resolution vs ResShift | CART-512-d30 · ImageNet SR 128->512 | PSNR: PSNR 24.65 vs 24.90; CLIPIQA 0.672 vs 0.603 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- DiffiT — DiffiT: Diffusion Vision Transformers for Image Generation (ECCV 2024) · better on 1 of 1 shared result
- ResShift-15 — ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS 2023) · 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
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.89)
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.89)
- A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation (ICLR 2025 · similarity 0.88)
- RandAR: Decoder-only Autoregressive Visual Generation in Random Orders (CVPR 2025 · similarity 0.88)
- JetFormer: An autoregressive generative model of raw images and text (ICLR 2025 · similarity 0.88)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · 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.
