Task-Balanced Distillation for Object Detection

08/05/2022
by   Ruining Tang, et al.
3

Mainstream object detectors are commonly constituted of two sub-tasks, including classification and regression tasks, implemented by two parallel heads. This classic design paradigm inevitably leads to inconsistent spatial distributions between classification score and localization quality (IOU). Therefore, this paper alleviates this misalignment in the view of knowledge distillation. First, we observe that the massive teacher achieves a higher proportion of harmonious predictions than the lightweight student. Based on this intriguing observation, a novel Harmony Score (HS) is devised to estimate the alignment of classification and regression qualities. HS models the relationship between two sub-tasks and is seen as prior knowledge to promote harmonious predictions for the student. Second, this spatial misalignment will result in inharmonious region selection when distilling features. To alleviate this problem, a novel Task-decoupled Feature Distillation (TFD) is proposed by flexibly balancing the contributions of classification and regression tasks. Eventually, HD and TFD constitute the proposed method, named Task-Balanced Distillation (TBD). Extensive experiments demonstrate the considerable potential and generalization of the proposed method. Specifically, when equipped with TBD, RetinaNet with ResNet-50 achieves 41.0 mAP under the COCO benchmark, outperforming the recent FGD and FRS.

READ FULL TEXT

page 1

page 3

page 5

page 6

page 7

page 8

page 9

page 13

research
11/15/2022

Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling

DETR is a novel end-to-end transformer architecture object detector, whi...
research
03/03/2021

General Instance Distillation for Object Detection

In recent years, knowledge distillation has been proved to be an effecti...
research
08/28/2023

Bridging Cross-task Protocol Inconsistency for Distillation in Dense Object Detection

Knowledge distillation (KD) has shown potential for learning compact mod...
research
10/27/2021

Beyond Classification: Knowledge Distillation using Multi-Object Impressions

Knowledge Distillation (KD) utilizes training data as a transfer set to ...
research
05/23/2022

PointDistiller: Structured Knowledge Distillation Towards Efficient and Compact 3D Detection

The remarkable breakthroughs in point cloud representation learning have...
research
04/03/2019

A Comprehensive Overhaul of Feature Distillation

We investigate the design aspects of feature distillation methods achiev...
research
01/04/2023

StereoDistill: Pick the Cream from LiDAR for Distilling Stereo-based 3D Object Detection

In this paper, we propose a cross-modal distillation method named Stereo...

Please sign up or login with your details

Forgot password? Click here to reset