HandsOff: Labeled Dataset Generation With No Additional Human Annotations

TL;DR

HandsOff is a framework that generates unlimited labeled synthetic datasets from very few initial images, eliminating the need for additional annotations and improving performance in various computer vision tasks.

Contribution

The paper introduces HandsOff, a novel method combining GAN inversion with dataset generation to produce labeled datasets with minimal initial data and no extra annotation effort.

Findings

Achieves state-of-the-art results in semantic segmentation, keypoint detection, and depth estimation.

Successfully generates diverse datasets in challenging domains like faces, cars, and urban scenes.

Addresses issues like the long-tail problem in dataset annotation.

Abstract

Recent work leverages the expressive power of generative adversarial networks (GANs) to generate labeled synthetic datasets. These dataset generation methods often require new annotations of synthetic images, which forces practitioners to seek out annotators, curate a set of synthetic images, and ensure the quality of generated labels. We introduce the HandsOff framework, a technique capable of producing an unlimited number of synthetic images and corresponding labels after being trained on less than 50 pre-existing labeled images. Our framework avoids the practical drawbacks of prior work by unifying the field of GAN inversion with dataset generation. We generate datasets with rich pixel-wise labels in multiple challenging domains such as faces, cars, full-body human poses, and urban driving scenes. Our method achieves state-of-the-art performance in semantic segmentation, keypoint detection, and depth estimation compared to prior dataset generation approaches and transfer learning baselines. We additionally showcase its ability to address broad challenges in model development which stem from fixed, hand-annotated datasets, such as the long-tail problem in semantic segmentation. Project page: austinxu87.github.io/handsoff.