AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

GarmentGPT: Compositional Garment Pattern Generation via Discrete Latent Tokenization

Fangsheng Weng, Junhao Chen, Xiang Li, Jie Qin, Hanzhong Guo, ShaochunHao, Xiaoguang Han

ICLR 2026 · Image generation › Autoregressive and latent generation · approach: Generative latent space tooling

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 changedWhereMeasuredEffect
add text input to image input (ablation)GarmentGPT (LLaVA-1.5) · GarmentCode datasetPanel Accuracy: 93.53% -> 95.62%Improved
add image input to text input (ablation)GarmentGPT (LLaVA-1.5) · GarmentCode datasetPanel Accuracy: 64.03% -> 95.62%Improved
replace AIpparel continuous regression with discrete tokensGarmentGPT · GarmentCode datasetPanel Accuracy: 78.92% -> 95.62%Improved
replace AIpparel continuous regression with discrete tokensGarmentGPT · GarmentCode datasetStitch Accuracy: 56.57% -> 81.84%Improved
scale Qwen backbone from 3B to 7B (ablation)GarmentGPT · GarmentCode datasetPanel Accuracy: 85.56% -> 90.31%Improved
scale Qwen backbone from 7B to 32B (ablation)GarmentGPT · GarmentCode datasetPanel Accuracy: 90.31% -> 91.05%No clear effect
use LLaVA-1.5 backbone instead of Qwen-7B (ablation)GarmentGPT · GarmentCode datasetPanel Accuracy: 90.31% -> 95.62%Improved
train on real-photo data and evaluate on real photosGarmentGPT vs AIpparel · Real-Garments BenchmarkPanel Accuracy: 38.76% -> 90.84%Improved
increase residual quantizers from 1 to 8 (ablation)RVQ-VAE · RVQ-VAE codec reconstructionTotal loss: 3.72 -> 0.08Improved
increase residual quantizers from 1 to 8 (ablation)RVQ-VAE · RVQ-VAE codec reconstructionCurve accuracy: 93.3 -> 99.8Improved
add multi-stage quality filtering to data pipelineRealGarment-1M pipeline · alignment acceptabilityacceptability: 64.3% -> 99.6%Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy