追寻面向视觉预训练的像素监督
In Pursuit of Pixel Supervision for Visual Pre-training
把数据策展（self-curation）当作与 MAE 预训练目标同等重要的一环：在 2B 网络图像上筛选空间结构信号更强的样本，可显著提升单目深度估计、前馈 3D 重建、分割等密集预测任务的表征质量，为视觉基础模型的预训练数据配方提供可复用思路。
原文依据 training on 2B web-crawled images with a self-curation strategy on masked autoencoder (MAE) can learn strong representations for dense prediction tasks
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论数据至关重要。在计算机视觉中，数据（或像素）是信息的主要来源，其中包含从低层属性到高层概念的各类信号。在大规模条件下，现代视觉系统的成功与数据如何被策展以服务于语义理解（例如 ImageNet）密切相关。近期空间智能与物理世界理解的趋势，进一步凸显了保留空间结构的真实世界信号的重要性，其价值超越了纯粹的语义信号。这促使我们转向策展能够更好地捕捉跨多样环境的空间接地信息的数据。在本工作中，我们证明：在 20 亿张网络爬取图像上，采用自策展策略对掩码自编码器（MAE）进行训练，可以为密集预测任务学习到强表征，同时保持简单、稳定与高效。我们的模型代号为“Pixio”，是一个增强型掩码自编码器（MAE），具有更具挑战性的预训练任务与能力更强的架构。Pixio 产生的密集表征在真实场景下广泛的密集预测任务上取得了有前景的结果，包括单目深度估计（例如 Depth Anything）、前馈三维重建（即 MapAnything）以及视觉分割（例如 SAM）。我们的结果表明，数据策展能够显著促进密集表征学习。
英文原文摘要 1371 字符
Data matters. In computer vision, data (or pixels) are the primary source of information containing signals that span from low-level attributes to high-level concepts. At scale, the success of modern vision systems has been closely tied to how data is curated for semantic understanding (e.g., ImageNet). Recent trends in spatial intelligence and physical world understanding further highlight the importance of real-world signals that preserve spatial structure, beyond purely semantic signals. This motivates a shift toward curating data that better captures spatially grounded information across diverse environments. In this work, we demonstrate that training on 2B web-crawled images with a self-curation strategy on masked autoencoder (MAE) can learn strong representations for dense prediction 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. Pixio yields dense representations achieving promising results across a wide range of dense prediction tasks in the wild, including monocular depth estimation (e.g., Depth Anything), feed-forward 3D reconstruction (i.e., MapAnything), and visual segmentation (e.g., SAM). Our results suggest that data curation can significantly contribute to dense representation learning.