DR-KFS: A Differentiable Visual Similarity Metric for 3D Shape Reconstruction

TL;DR

This paper presents a differentiable visual similarity metric for 3D shape reconstruction that improves the quality of reconstructions by focusing on perceptual similarity rather than pixel-wise errors.

Contribution

The authors introduce a novel differentiable visual similarity metric based on multi-view image features, which can replace traditional distortion metrics in 3D reconstruction networks.

Findings

Improves reconstruction quality in terms of structural fidelity and visual perception.

Enhances 3D shape retrieval and classification performance.

Validated through perceptual studies and shape metrics.

Abstract

We introduce a differential visual similarity metric to train deep neural networks for 3D reconstruction, aimed at improving reconstruction quality. The metric compares two 3D shapes by measuring distances between multi-view images differentiably rendered from the shapes. Importantly, the image-space distance is also differentiable and measures visual similarity, rather than pixel-wise distortion. Specifically, the similarity is defined by mean-squared errors over HardNet features computed from probabilistic keypoint maps of the compared images. Our differential visual shape similarity metric can be easily plugged into various 3D reconstruction networks, replacing their distortion-based losses, such as Chamfer or Earth Mover distances, so as to optimize the network weights to produce reconstructions with better structural fidelity and visual quality. We demonstrate this both objectively, using well-known shape metrics for retrieval and classification tasks that are independent from our new metric, and subjectively through a perceptual study.