Recurrent 3D Attentional Networks for End-to-End Active Object Recognition

TL;DR

This paper introduces a recurrent 3D attentional network that actively selects views for efficient and accurate 3D object recognition using depth data, outperforming previous methods in speed and accuracy.

Contribution

It presents a novel end-to-end recurrent 3D attentional model with a differentiable view selection mechanism, improving over reinforcement learning approaches.

Findings

Achieves state-of-the-art next-best-view performance.

Faster convergence than reinforcement learning methods.

Effective recognition using only depth input.

Abstract

Active vision is inherently attention-driven: The agent actively selects views to attend in order to fast achieve the vision task while improving its internal representation of the scene being observed. Inspired by the recent success of attention-based models in 2D vision tasks based on single RGB images, we propose to address the multi-view depth-based active object recognition using attention mechanism, through developing an end-to-end recurrent 3D attentional network. The architecture takes advantage of a recurrent neural network (RNN) to store and update an internal representation. Our model, trained with 3D shape datasets, is able to iteratively attend to the best views targeting an object of interest for recognizing it. To realize 3D view selection, we derive a 3D spatial transformer network which is differentiable for training with backpropagation, achieving much faster convergence than the reinforcement learning employed by most existing attention-based models. Experiments show that our method, with only depth input, achieves state-of-the-art next-best-view performance in time efficiency and recognition accuracy.