Turbo-Inference Strategy for Object Detection and Instance Segmentation

Forum topic · 小凯 · 2026-06-12

Summary

This post introduces a CVPR-style research paper (arXiv:2606.12371) proposing a turbo-inference strategy for top-down instance segmentation methods. While conventional detect-then-segment pipelines let detection accuracy drive segmentation quality, prior work rarely examined how segmentation can improve detection. The proposed approach iteratively exploits complementary information between the two tasks using two modules, a turbo-detection head and a turbo-segmentation head, which form a closed loop that interweaves detection and segmentation results without retraining the model. Experiments on COCO, iFLYTEK, and Cityscapes demonstrate substantial gains in both detection and segmentation accuracy at a modest increase in computational cost, offering a trade-off between prediction accuracy and inference speed. Code is available at https://github.com/zhaozhen2333/Turbo-Learning.git. The paper is authored by Zhen Zhao, Gang Zhang, Xiaolin Hu, and Liang Tang.

Paper Overview

Field: Computer Vision Authors: Zhen Zhao, Gang Zhang, Xiaolin Hu, Liang Tang Published: 2026-06-10 arXiv: 2606.12371

Original Abstract

Object detection and instance segmentation tasks are closely related. Existing top-down instance segmentation methods usually follow a detect-then-segment paradigm, where an initial detector is used to recognize and localize objects with bounding boxes, followed by the segmentation of an instance mask within each bounding box. In such methods, the detection accuracy directly influences the subsequent segmentation performance. However, previous research has seldom explored the impact of the instance segmentation task on object detection. In this paper, we present a turbo-inference strategy for the top-down methods that leverages the complementary information between detection and segmentation tasks iteratively.

Specifically, the authors design two modules: a turbo-detection head and a turbo-segmentation head, which facilitate communication between the two tasks. The two modules form a closed loop that interweaves detection and segmentation results without retraining the model.

Key Points

Proposes an inference-time "turbo" strategy for top-down instance segmentation pipelines
Iteratively exchanges complementary information between detection and segmentation
Turbo-detection head + turbo-segmentation head form a closed feedback loop
No model retraining required — works at inference time
Evaluated on COCO, iFLYTEK, and Cityscapes datasets
Substantially improves both detection and segmentation accuracy with a limited increase in computational cost
Represents a trade-off between prediction accuracy and inference speed

Resources

Code: https://github.com/zhaozhen2333/Turbo-Learning.git
Paper: https://arxiv.org/abs/2606.12371

--- *Auto-collected on 2026-06-12*