Label-Free Synthetic Pretraining of Object Detectors

TL;DR

This paper introduces SOLID, a synthetic pretraining method for object detectors using optimized scene arrangement and instance detection tasks, achieving competitive performance with less computational cost.

Contribution

The paper presents a novel synthetic pretraining approach that does not require semantic labels and leverages diverse 3D models for effective object detector pretraining.

Findings

Synthetic pretraining with SOLID is competitive with real data pretraining.

Optimized scene arrangement enhances synthetic data effectiveness.

Pretraining on rendered images reduces computational resources.

Abstract

We propose a new approach, Synthetic Optimized Layout with Instance Detection (SOLID), to pretrain object detectors with synthetic images. Our "SOLID" approach consists of two main components: (1) generating synthetic images using a collection of unlabelled 3D models with optimized scene arrangement; (2) pretraining an object detector on "instance detection" task - given a query image depicting an object, detecting all instances of the exact same object in a target image. Our approach does not need any semantic labels for pretraining and allows the use of arbitrary, diverse 3D models. Experiments on COCO show that with optimized data generation and a proper pretraining task, synthetic data can be highly effective data for pretraining object detectors. In particular, pretraining on rendered images achieves performance competitive with pretraining on real images while using significantly less computing resources. Code is available at https://github.com/princeton-vl/SOLID.