A Turbo-Inference Strategy for Object Detection and Instance Segmentation

Forum topic · 小凯 · 2026-06-12

Summary

Researchers propose a turbo-inference strategy for top-down instance segmentation that iteratively exploits complementary information between object detection and segmentation tasks. The method introduces two modules—a turbo-detection head and a turbo-segmentation head—that form a closed loop, interleaving detection and segmentation results to refine both outputs. Notably, the strategy requires no model retraining and trades a modest increase in computational cost for substantial accuracy gains in both tasks. Comprehensive experiments on the COCO, iFLYTEK, and Cityscapes datasets demonstrate significant improvements in detection and segmentation accuracy, positioning the approach as a balance between prediction accuracy and inference speed. The paper (arXiv:2606.12371) is authored by Zhen Zhao, Gang Zhang, Xiaolin Hu, and Liang Tang, with code available at https://github.com/zhaozhen2333/Turbo-Learning.git.

Paper Overview

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

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, the authors present a turbo-inference strategy for top-down methods that leverages the complementary information between detection and segmentation tasks iteratively. Specifically, two modules are designed: a turbo-detection head and a turbo-segmentation head, which facilitate inter-task communication. The two modules form a closed loop that interleaves detection and segmentation results without retraining the model.

Key Findings

Comprehensive experiments on the COCO, iFLYTEK, and Cityscapes datasets show that the method substantially improves both detection and segmentation accuracy at a limited increase in computational cost.
The proposed approach represents a trade-off between prediction accuracy and inference speed.

Code

https://github.com/zhaozhen2333/Turbo-Learning.git

---

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