Geometric Reciprocity: Unlocking Self-Supervision for Stereoscopic Video Generation

Jul 16, 2026

Upload by Ezra

Monocular-to-stereo conversion synthesizes immersive 3D content by transferring depth-image-based rendering (DIBR) across views. Existing stereo inpainting methods become bottlenecks because disocclusions form structured regions that generic inpainting struggles to fill. Training approaches need scarce stereo pairs or synthetic data with domain gaps, while training-free methods lack stereo-specific geometric priors. This work introduces a self-supervised framework using cycle consistency grounded in a Geometric Reciprocity Theorem, enabling analytical test-time disocclusion masks from monocular images and improving train-test consistency.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

Introduction

DIBR framework and stereo inpainting bottleneck

Data scarcity and limitations of existing methods

Proposed self-supervised approach and cycle consistency

Geometric Reciprocity Theorem (GRT)

Geometric Reciprocity: Unlocking Self-Supervision for Stereoscopic Video Generation Jingyi Lu 1 Kai Han 1 arXiv :2607 .05354v 1 [ cs .CV] 6 Jul 2026 Abstract Monocular-to-stereo conversion synthesizes stereoscopic content from 2D videos for immersive 3D experiences. In modern DepthImage-Based Rendering (DIBR) approaches, stereo inpainting of disocclusions is the critical bottleneck. Training-based methods achieve superior quality but rely on scarce stereo pairs or synthetic data with domain gaps. We address this through the first self-supervised framework learning from monocular videos via cycle consistency. Our key contribution is the Geometric Reciprocity Theorem (GRT) : under the nearest-neighbor DIBR formulation, the disocclusion mask when synthesizing a target view equals the mask of pixels lost when warping back from target to source, enabling analytical computation of test-time disocclusion masks directly from monocular images. This yields train-test consistency for the stated warping formulation, supporting self-supervised learning from unlimited monocular videos and substantial improvements over training-free and supervised state-of-the-art methods. Project page: [https://visual-ai.github.io/grt/](https://visual-ai.github.io/grt/) 1. Introduction The demand for immersive 3D experiences on VR/AR devices, 3D cinema, and stereoscopic displays has made stereoscopic video generation a fundamental problem. The task is to convert monocular videos to stereoscopic format by synthesizing one view from the other (conventionally right-eye from left-eye, or vice versa) . Modern approaches predominantly adopt the Depth-ImageBased Rendering (DIBR) framework (Wang et al., 2024 ; Shi et al., 2024 ; Dai et al., 2024 ; Zhao et al., 2024 ; Huang et al., 1Visual AI Lab, The University of Hong Kong, Hong Kong, China. Correspondence to: Kai Han <kaihanx@hku.hk> . Proceedings of the 43 rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026 . Copyright 2026 by the author(s) . 2025 ; Shvetsova et al., 2026), which decomposes the problem into three sequential stages: estimating depth from the input frame, warping the image to synthesize an initial target view, and inpainting the resulting disocclusions. Thesedisocclusions are regions newly visible in the target view that were occluded in the source, creating geometrically structured missing patterns that general-purpose inpainting methods cannot handle (see Section A7) . This domain gap has made the stereo inpainting stage the critical bottleneck. Due to the scarcity of training data, recent training-free methods (Wang et al., 2024 ; Dai et al., 2024) manipulate pretrained diffusion models for zero-shot stereo inpainting, but lack stereo-specific geometric priors and produce inferior results. Training-based approaches (Zhao et al., 2024 ; Huang et al., 2025 ; Shvetsova et al., 2026) learn these priors yet face a fundamental challenge in training data quality and availability. Existing methods either rely on scarce and often proprietary real stereo pairs with error-prone stereo matching for disocclusion identification (Zhao et al., 2024 ; Shvetsova et al., 2026), or resort to synthetic data that suffers from inevitable domain gaps (Huang et al., 2025) (see Section A8 for more details) . To address the data bottleneck, we propose the first selfsupervised framework that learns stereo inpainting from monocular videos alone, eliminating the need for stereo pairs or synthetic data. Our approach exploits the inherent bidirectional symmetry of stereo view relationships, enabling right-left-right cycle consistency where generating the left view from right through DIBR and then reconstructing the right from the synthesized left should recover the original input. While cycle consistency provides self-supervision in principle, naive implementation requires multiple sequential model inferences and backpropagation through nondifferentiable warping

FAQs

What problem does the method address in monocular-to-stereo video generation?

The method targets stereo inpainting of disocclusions in DIBR, which forms geometrically structured missing regions and is the critical bottleneck. It aims to learn this stage without relying on stereo pairs or synthetic data.

Why do training-based and training-free approaches struggle?

Training-based approaches depend on scarce high-quality stereo data or synthetic data that introduces domain gaps. Training-free approaches using pretrained diffusion models lack stereo-specific geometric priors, producing inferior results.

How does the Geometric Reciprocity Theorem enable self-supervised learning?

Under a nearest-neighbor DIBR formulation, the theorem states that the disocclusion mask for synthesizing a target view equals the pixels lost when warping back from target to source. This allows computing train-time disocclusion masks analytically from monocular images using only depth estimates, avoiding explicit cycle synthesis.

Jul 16, 2026

Upload by Ezra