Bamboo: Building Mega-Scale Vision Dataset Continually with Human-Machine Synergy

TL;DR

The paper addresses the inefficiency of annotating mega-scale vision datasets with sample-agnostic labeling. It proposes a human-machine synergy-based active-learning framework designed to operate under realistic annotation conditions with substantial out-of-distribution data. Bamboo is built as a mega-scale dataset containing 69M image classification annotations across 119K categories and 28M bounding-box annotations across 809 categories, organized by a hierarchical taxonomy from multiple knowledge bases. Pre-training models on Bamboo yields 6.2% gains in classification and 2.1% gains in detection compared with baselines such as ImageNet22K and Object365. The work demonstrates a scalable, high-quality data construction approach with strong downstream benefits, establishing a framework and dataset that can guide future mega-scale vision data collection.

Abstract

Large-scale datasets play a vital role in computer vision. But current datasets are annotated blindly without differentiation to samples, making the data collection inefficient and unscalable. The open question is how to build a mega-scale dataset actively. Although advanced active learning algorithms might be the answer, we experimentally found that they are lame in the realistic annotation scenario where out-of-distribution data is extensive. This work thus proposes a novel active learning framework for realistic dataset annotation. Equipped with this framework, we build a high-quality vision dataset -- Bamboo, which consists of 69M image classification annotations with 119K categories and 28M object bounding box annotations with 809 categories. We organize these categories by a hierarchical taxonomy integrated from several knowledge bases. The classification annotations are four times larger than ImageNet22K, and that of detection is three times larger than Object365. Compared to ImageNet22K and Objects365, models pre-trained on Bamboo achieve superior performance among various downstream tasks (6.2% gains on classification and 2.1% gains on detection). We believe our active learning framework and Bamboo are essential for future work.