Learning Deep Object Detectors from 3D Models

TL;DR

This paper demonstrates that synthetic training data generated from 3D CAD models can significantly improve deep object detection, especially in limited data or domain shift scenarios, revealing the DCNN's invariance to missing low-level cues.

Contribution

It introduces a synthetic data augmentation approach using 3D CAD models for training deep object detectors, showing its effectiveness over previous methods.

Findings

Synthetic data improves detection performance in few-shot learning.

Pretraining on ImageNet benefits from simulated low-level cues.

Method outperforms previous approaches on PASCAL VOC2007 and Office benchmarks.

Abstract

Crowdsourced 3D CAD models are becoming easily accessible online, and can potentially generate an infinite number of training images for almost any object category.We show that augmenting the training data of contemporary Deep Convolutional Neural Net (DCNN) models with such synthetic data can be effective, especially when real training data is limited or not well matched to the target domain. Most freely available CAD models capture 3D shape but are often missing other low level cues, such as realistic object texture, pose, or background. In a detailed analysis, we use synthetic CAD-rendered images to probe the ability of DCNN to learn without these cues, with surprising findings. In particular, we show that when the DCNN is fine-tuned on the target detection task, it exhibits a large degree of invariance to missing low-level cues, but, when pretrained on generic ImageNet classification, it learns better when the low-level cues are simulated. We show that our synthetic DCNN training approach significantly outperforms previous methods on the PASCAL VOC2007 dataset when learning in the few-shot scenario and improves performance in a domain shift scenario on the Office benchmark.