Zongxi Yu*, Xiaolong Qian*, Shaohua Gao, Qi Jiang, Yao Gao, Kaiwei Wang†,Kailun Yang†

High-fidelity, compact RGBD imaging faces a dual challenge: conventional compact optics struggle with sharpness across the entire depth-of-field, while software-only Monocular Depth Estimation (MDE) is an ill-posed problem reliant on unreliable priors. To address this, we introduce the Bionic Monocentric Imaging (BMI) framework, a holistic co-design centered around a novel bio-inspired, all-spherical monocentric lens. This optical design naturally encodes depth into its depth-varying Point Spread Functions (PSFs) without requiring complex elements. We co-designed a rigorous simulation pipeline with a dual-head reconstruction network that jointly recovers a high-fidelity All-in-Focus (AiF) image and a precise depth map from a single coded capture5.Extensive experiments validate our state-of-the-art performance6. Our method achieves an Abs Rel of 0.026 and an RMSE of 0.130 in depth estimation 7, and an SSIM of 0.960 with an LPIPS of 0.082 for image restoration 8, demonstrating a superior balance between image fidelity and depth accuracy.

The source code will be made publicly available after the paper is accepted.