Pose Adaptive Dual Mixup for Few-Shot Single-View 3D Reconstruction

TL;DR

This paper introduces PADMix, a pose adaptive dual mixup method that enhances few-shot single-view 3D reconstruction by improving data augmentation and pose invariance, leading to state-of-the-art results on ShapeNet and Pix3D.

Contribution

The paper proposes a novel pose adaptive dual mixup technique with a two-stage training process for improved few-shot 3D reconstruction from single images.

Findings

Outperforms previous methods on ShapeNet dataset

Sets new benchmarks on Pix3D dataset

Improves shape prediction accuracy in few-shot settings

Abstract

We present a pose adaptive few-shot learning procedure and a two-stage data interpolation regularization, termed Pose Adaptive Dual Mixup (PADMix), for single-image 3D reconstruction. While augmentations via interpolating feature-label pairs are effective in classification tasks, they fall short in shape predictions potentially due to inconsistencies between interpolated products of two images and volumes when rendering viewpoints are unknown. PADMix targets this issue with two sets of mixup procedures performed sequentially. We first perform an input mixup which, combined with a pose adaptive learning procedure, is helpful in learning 2D feature extraction and pose adaptive latent encoding. The stagewise training allows us to build upon the pose invariant representations to perform a follow-up latent mixup under one-to-one correspondences between features and ground-truth volumes. PADMix significantly outperforms previous literature on few-shot settings over the ShapeNet dataset and sets new benchmarks on the more challenging real-world Pix3D dataset.