PXDepth: Pixel-Space Modeling for Structure Preserving Monocular Depth Estimation

PXDepth：用于结构保持单目深度估计的像素空间建模

Abstract: Recent monocular depth estimators achieve strong zero-shot generalization, yet often struggle to preserve fine-grained structures and object boundaries. We attribute this limitation to the prevalent combination of large-patch ViT encoders and convolutional decoders, as coarse tokenization can weaken pixel-level cues that upsampling cannot fully recover.

摘要： 近期的单目深度估计器虽然实现了强大的零样本泛化能力，但在保持细粒度结构和物体边界方面往往表现欠佳。我们将这一局限性归因于目前主流的“大补丁 ViT 编码器 + 卷积解码器”组合，因为粗糙的标记化（tokenization）会削弱像素级线索，而上采样过程无法完全恢复这些信息。

To address this issue, we propose PXDepth, a discriminative monocular depth model that separates global context modeling from pixel-level depth prediction. Specifically, a large-patch ViT captures global scene context, while a pixel-space predictor composed of Context-Modulated Pixel Transformer blocks maintains high-resolution spatial representations throughout depth estimation.

为了解决这一问题，我们提出了 PXDepth，这是一种判别式单目深度模型，它将全局上下文建模与像素级深度预测分离开来。具体而言，模型利用大补丁 ViT 捕捉全局场景上下文，同时由“上下文调制像素 Transformer”块组成的像素空间预测器在整个深度估计过程中保持高分辨率的空间表征。

This design preserves fine structures and sharp boundaries without sacrificing global depth consistency. Across diverse zero-shot benchmarks, PXDepth combines faithful local geometry with competitive global depth accuracy while remaining efficient at inference. Our code and model are available at this https URL.

这种设计在不牺牲全局深度一致性的前提下，保留了精细的结构和清晰的边界。在多种零样本基准测试中，PXDepth 不仅实现了精准的局部几何还原，还具备极具竞争力的全局深度精度，同时保持了高效的推理速度。我们的代码和模型已在链接中提供。

---

Paper Details:

• Authors: Zhiyuan Yuan, Guanying Chen, Lingteng Qiu, Ruimao Zhang, Shuguang Cui, Xiaochun Cao
• Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR)
• arXiv ID: 2608.16984

论文详情：

• 作者： Zhiyuan Yuan, Guanying Chen, Lingteng Qiu, Ruimao Zhang, Shuguang Cui, Xiaochun Cao
• 学科分类： 计算机视觉与模式识别 (cs.CV)；人工智能 (cs.AI)；图形学 (cs.GR)
• arXiv 编号： 2608.16984