Curiosity-driven 3D Object Detection Without Labels

TL;DR

This paper introduces a self-supervised method for 6-DOF 3D object detection from images, using a neural network with explicit scene parameterization and curiosity-driven training to overcome ambiguities without requiring labeled data.

Contribution

It proposes a novel curiosity-driven training approach that enables 3D object detection without 6-DOF labels, utilizing differentiable rendering and an auxiliary network for exploration.

Findings

Achieves 3D object detection from unlabeled images.

Demonstrates the effectiveness of curiosity-driven exploration.

Analyzes limitations of analysis-by-synthesis losses.

Abstract

In this paper we set out to solve the task of 6-DOF 3D object detection from 2D images, where the only supervision is a geometric representation of the objects we aim to find. In doing so, we remove the need for 6-DOF labels (i.e., position, orientation etc.), allowing our network to be trained on unlabeled images in a self-supervised manner. We achieve this through a neural network which learns an explicit scene parameterization which is subsequently passed into a differentiable renderer. We analyze why analysis-by-synthesis-like losses for supervision of 3D scene structure using differentiable rendering is not practical, as it almost always gets stuck in local minima of visual ambiguities. This can be overcome by a novel form of training, where an additional network is employed to steer the optimization itself to explore the entire parameter space i.e., to be curious, and hence, to resolve those ambiguities and find workable minima.