From Pixels, Without Pre-training: Joint Generative and Self-Supervised Representation Learning in One Model

2026-10-05

AI Summary

This study addresses the limitation of existing image generation methods that rely on labels or pretrained models, which hinders unsupervised joint generation and representation learning. To this end, it proposes SCION, a self-conditioned generative architecture that unifies contrastive learning and flow matching objectives via a single-pixel spatial encoder. By integrating coarse-grained semantic invariance with fine-grained spatial details and introducing gradient norm balancing alongside stop-gradient mechanisms to mitigate multi-task conflicts, the framework enables effective end-to-end joint optimization. The proposed method achieves state-of-the-art performance on ImageNet, with the JiT-B and JiT-L configurations yielding FID scores of 8.92 and 5.89, respectively, without guidance. These results significantly outperform prior approaches that depend on pretrained alignment.

Abstract

Strong image generation models are conditioned on class labels, aligned to frozen pretrained encoders, or built on separately trained autoencoders. While effective, generation then depends on supervision or pretraining: labels must be annotated, and encoders or autoencoders pretrained for the target domain. We study joint generative and self-supervised representation learning in a single model, enabling self-conditioned generation without labels or pretrained models. This is challenging because the objectives are mismatched: contrastive learning consumes clean augmented views and favors coarse, invariant semantics, while flow matching consumes noisy images and must preserve the fine detail and spatial layout that contrastive learning discards. We propose SCION (Self-conditioned Generation on Self-supervised representation), whose core is a single pixel-space encoder conditioned on the flow timestep and an embedding. For representation learning, this conditioning embedding is a learned global vector shared across images, with the encoder's [CLS] token yielding the semantic representation trained by the contrastive loss. For generative training, the conditioning embedding is the image's own [CLS] representation, while patch tokens pass through a decoder to predict the image. To sample without a reference image at inference, we jointly learn a prior over the embedding. Gradient-norm balancing and stop-gradient mechanisms enable joint optimization in one run. SCION is self-supervised and self-contained, with no labels or pretrained models. On ImageNet 256x256, with the JiT-B recipe and no representation guidance, SCION reaches 8.92 FID, surpassing class-unconditional iREPA, which aligns to pretrained DINOv2 (46.44), and RCG, which conditions on it (14.27). With JiT-L, SCION achieves 5.89 FID without guidance and 3.47 with representation guidance, outperforming RCG with the ADM recipe (6.24).

Problem

• image generation
• self-supervised representation learning
• joint learning
• contrastive learning
• flow matching

Innovation

• Self-supervised representation learning
• Joint generative model
• Flow matching
• Contrastive learning
• Self-conditioned generation