In Pursuit of Pixel Supervision for Visual Pre-training
This work addresses the challenge of efficient and stable visual pre-training for downstream tasks in computer vision, offering a competitive alternative to latent-space methods, though it is incremental as it builds on existing autoencoder paradigms.
The authors tackled the problem of learning visual representations from pixels using autoencoder-based self-supervised learning, demonstrating that their enhanced masked autoencoder model, Pixio, trained on 2B web-crawled images, performs competitively or matches DINOv3 across tasks like monocular depth estimation and semantic segmentation.
At the most basic level, pixels are the source of the visual information through which we perceive the world. Pixels contain information at all levels, ranging from low-level attributes to high-level concepts. Autoencoders represent a classical and long-standing paradigm for learning representations from pixels or other raw inputs. In this work, we demonstrate that autoencoder-based self-supervised learning remains competitive today and can produce strong representations for downstream tasks, while remaining simple, stable, and efficient. Our model, codenamed "Pixio", is an enhanced masked autoencoder (MAE) with more challenging pre-training tasks and more capable architectures. The model is trained on 2B web-crawled images with a self-curation strategy with minimal human curation. Pixio performs competitively across a wide range of downstream tasks in the wild, including monocular depth estimation (e.g., Depth Anything), feed-forward 3D reconstruction (i.e., MapAnything), semantic segmentation, and robot learning, outperforming or matching DINOv3 trained at similar scales. Our results suggest that pixel-space self-supervised learning can serve as a promising alternative and a complement to latent-space approaches.