DAP: Detection-Aware Pre-training with Weak Supervision

TL;DR

This paper introduces Detection-Aware Pre-training (DAP), a method that uses weakly-labeled datasets like ImageNet to pre-train models specifically for object detection, improving efficiency and accuracy especially with limited data.

Contribution

The novel approach transforms classification datasets into detection datasets using weak supervision, enabling detection-aware pre-training from classification data.

Findings

DAP outperforms traditional classification pre-training in detection tasks.

DAP improves sample efficiency and convergence speed.

Significant accuracy gains with small training datasets.

Abstract

This paper presents a detection-aware pre-training (DAP) approach, which leverages only weakly-labeled classification-style datasets (e.g., ImageNet) for pre-training, but is specifically tailored to benefit object detection tasks. In contrast to the widely used image classification-based pre-training (e.g., on ImageNet), which does not include any location-related training tasks, we transform a classification dataset into a detection dataset through a weakly supervised object localization method based on Class Activation Maps to directly pre-train a detector, making the pre-trained model location-aware and capable of predicting bounding boxes. We show that DAP can outperform the traditional classification pre-training in terms of both sample efficiency and convergence speed in downstream detection tasks including VOC and COCO. In particular, DAP boosts the detection accuracy by a large margin when the number of examples in the downstream task is small.