Deep Structured Instance Graph for Distilling Object Detectors

TL;DR

This paper introduces a graph-based knowledge distillation method for object detectors that encodes instance relations to improve student model performance, achieving state-of-the-art results on COCO.

Contribution

It proposes a novel graph-structured knowledge representation for detector distillation, capturing local and global relations to enhance transfer learning.

Findings

Achieves state-of-the-art COCO detection results with various student-teacher pairs.

Outperforms baseline models in object detection and instance segmentation.

Demonstrates robustness across different detector architectures.

Abstract

Effectively structuring deep knowledge plays a pivotal role in transfer from teacher to student, especially in semantic vision tasks. In this paper, we present a simple knowledge structure to exploit and encode information inside the detection system to facilitate detector knowledge distillation. Specifically, aiming at solving the feature imbalance problem while further excavating the missing relation inside semantic instances, we design a graph whose nodes correspond to instance proposal-level features and edges represent the relation between nodes. To further refine this graph, we design an adaptive background loss weight to reduce node noise and background samples mining to prune trivial edges. We transfer the entire graph as encoded knowledge representation from teacher to student, capturing local and global information simultaneously. We achieve new state-of-the-art results on the challenging COCO object detection task with diverse student-teacher pairs on both one- and two-stage detectors. We also experiment with instance segmentation to demonstrate robustness of our method. It is notable that distilled Faster R-CNN with ResNet18-FPN and ResNet50-FPN yields 38.68 and 41.82 Box AP respectively on the COCO benchmark, Faster R-CNN with ResNet101-FPN significantly achieves 43.38 AP, which outperforms ResNet152-FPN teacher about 0.7 AP. Code: https://github.com/dvlab-research/Dsig.