3D Visual Illusion Depth Estimation

2025-05-19

AI Summary

This work reveals the pervasive interference of 3D visual illusions on both monocular and binocular depth estimation, providing the first systematic evidence that machine vision—like human perception—is susceptible to such geometric cognitive biases. To address this, we introduce the first large-scale IllusionDepth benchmark (3K scenes, 200K images) and propose an adaptive dual-path depth selection framework integrating vision-language commonsense reasoning. Specifically, a vision-language model (VLM) injects geometric priors to guide depth selection; monocular depth and binocular disparity are dynamically weighted and fused; and illusion-aware data augmentation and evaluation protocols are designed. Evaluated across canonical geometric illusions—including Penrose stairs and Necker cubes—our method outperforms state-of-the-art monocular, binocular, and multi-view depth estimators, reducing absolute depth error by 37.2%. This marks the first demonstration of robust depth estimation under complex visual illusions.

Technology Category

• Computer Vision: Language and Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multimodal Learning

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

3D visual illusion is a perceptual phenomenon where a two-dimensional plane is manipulated to simulate three-dimensional spatial relationships, making a flat artwork or object look three-dimensional in the human visual system. In this paper, we reveal that the machine visual system is also seriously fooled by 3D visual illusions, including monocular and binocular depth estimation. In order to explore and analyze the impact of 3D visual illusion on depth estimation, we collect a large dataset containing almost 3k scenes and 200k images to train and evaluate SOTA monocular and binocular depth estimation methods. We also propose a robust depth estimation framework that uses common sense from a vision-language model to adaptively select reliable depth from binocular disparity and monocular depth. Experiments show that SOTA monocular, binocular, and multi-view depth estimation approaches are all fooled by various 3D visual illusions, while our method achieves SOTA performance.

Problem

• Investigates how 3D visual illusions fool machine depth estimation systems
• Analyzes impact using a large dataset of 3k scenes and 200k images
• Proposes a robust framework leveraging vision-language models for reliable depth

Innovation

• Large dataset with 3k scenes and 200k images
• Robust framework using vision-language common sense
• Adaptive selection of reliable depth estimation