GEAR: Guided End-to-End AutoRegression for Image Synthesis
2026-06-30 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created a new method called GEAR to train two parts of an image generation model together, rather than separately like before. They found a clever way to let the ‘tokenizer’ learn based on feedback from the ‘generator’, even though the usual signals don’t pass through easily. This helps the model make better and faster image predictions, improving training speed and image quality on standard tests. Their approach also works well with different types of tokenizers and for turning text into images.
Visual generative modelsTokenizerVector quantization (VQ)Autoregressive (AR) modelsRepresentation alignmentNext-token predictionDINOv2ImageNetgFIDText-to-image generation
Authors
Bin Lin, Zheyuan Liu, Chenguo Lin, Sixiang Chen, Yunyang Ge, Yunlong Lin, Jianwei Zhang, Miles Yang, Zhao Zhong, Liefeng Bo, Li Yuan
Abstract
Visual generative models are typically trained in two stages. A tokenizer is first trained for reconstruction and then frozen, after which a generator is trained on its discrete indices or continuous latents. This decoupling leaves the tokenizer unaware of what the generator finds easy to model. We present GEAR (Guided End-to-end AutoRegression), which trains a vector-quantized (VQ) tokenizer and an autoregressive (AR) generator jointly and end-to-end, guided by representation alignment. The key obstacle is that the VQ index fed to the AR model is non-differentiable, so gradients cannot reach the tokenizer, and a straight-through estimator collapses. GEAR resolves this with a dual read-out of the codebook assignment. A hard, one-hot branch trains the AR with next-token prediction, while a differentiable soft branch carries a representation-alignment loss that flows back to guide only the tokenizer. The AR model thereby steers its tokenizer toward an index distribution it can predict more easily. This shifts the alignment burden from the tokenizer to the AR: the tokenizer's own features become less DINOv2-like while the AR's become more so, the opposite of diffusion-side recipes that make the latent itself semantic. GEAR speeds up ImageNet gFID convergence by up to 10x relative to the strong LlamaGen-REPA baseline, learns markedly better patch-level and spatially-coherent features, and generalizes across quantizers (VQVAE, LFQ, IBQ) and to text-to-image generation.