OTA: Optimal Transport Assignment for Object Detection

By Zheng Ge, Songtao Liu, Zeming Li, Osamu Yoshie, Jian‐Yuan Sun. 2021

529 citations recorded by the source corpus.

Recent advances in label assignment in object detection mainly seek to independently define positive/negative training samples for each ground-truth (gt) object. In this paper, we innovatively revisit the label assignment from a global perspective and propose to formulate the assigning procedure as an Optimal Transport (OT) problem – a well-studied topic in Optimization Theory. Concretely, we define the unit transportation cost between each demander (anchor) and supplier (gt) pair as the weighted summation of their classification and regression losses. After formulation, finding the best assignment solution is converted to solve the optimal transport plan at minimal transportation costs, which can be solved via Sinkhorn-Knopp Iteration. On COCO, a single FCOS-ResNet-50 detector equipped with Optimal Transport Assignment (OTA) can reach 40.7% mAP under 1× scheduler, outperforming all other existing assigning methods. Extensive experiments conducted on COCO and CrowdHuman further validate the effectiveness of our proposed OTA, especially its superiority in crowd scenarios. The code is available at https://github.com/Megvii-BaseDetection/OTA.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Co-Transport for Class-Incremental Learning (2021, neutral): “model fusion [37], domain adaptation [7], object detection [10], model reuse [56, 57], and generative models [2, 4].”
YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications (2022, background): “SimOTA [7] is a simplified version of OTA [6], which reduces additional hyperparameters and maintains the performance.”
YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors (2022, methodology): “However, in recent years, if we take object detection as an example, researchers often use the quality and distribution of prediction output by the network, and then consider together with the ground truth to use some calculation and optimization methods to generate a reliable soft label [61, 8, 36,”
Oriented RepPoints for Aerial Object Detection (2021, methodology): “Some recent samples assignment methods in general object detection, such as the ATSS [48], FreeAn-chor [49], PAA [13] and OTA [6], employ a learning-to-match optimization strategy [50] to choose the high-quality samples.”
YOLOX: Exceeding YOLO Series in 2021 (2021, methodology): “…performance with 48.2% AP on COCO at 13.7 ms. 1 Nevertheless, over the past two years, the major advances in object detection academia have focused on anchor-free detectors [29, 40, 14], advanced label assignment strategies [37, 36, 12, 41, 22, 4], and end-to-end (NMS-free) detectors [2, 32, 39].”
Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark (2022, unknown): “Dataset 1 2 3 4 6 8 baseline AI-TOD-v2 20.5 21.4 20.9 21.0 20.9 20.7 11.1 VisDrone2019 23.2 23.2 23.4 23.0 22.9 23.0 22.3 DOTA-v2.”
YOLOv12: Attention-Centric Real-Time Object Detectors (2025, unknown): “Although improvements in YOLO have focused on areas such as loss functions [8, 35, 43, 44, 48, 67, 68], label assignment [22, 23, 34, 59, 69], network architecture design has remained a critical research priority [24, 28, 32, 57, 58].”
A Comprehensive Review of YOLO: From YOLOv1 to YOLOv8 and Beyond (2023, methodology): “In [84], it was shown that the ground truth label assignment could have ambiguities when the boxes of multiple objects overlap and formulate the assigning procedure as an Optimal Transport (OT) problem.”
DAMO-YOLO : A Report on Real-Time Object Detection Design (2022, methodology): “In addition, the dynamic label assignment, such as OTA [8] and TOOD [7], is widely acclaimed and achieves signiﬁcant improvement compared to the static label assignment [43].”
YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information (2024, background): “In addition, some researchers have tried to develop more general objective functions, such as loss function [5, 45, 46, 50, 77, 78], label assignment [10, 12, 33, 67, 79] and auxiliary supervision [18, 20, 24, 28, 29, 51, 54, 68, 76].”
DiffusionDet: Diffusion Model for Object Detection (2022, methodology): “We assign multiple predictions to each ground truth with the optimal transport approach [25, 26].”
TOOD: Task-aligned One-stage Object Detection (2021, methodology): “We use different values of m in [5, 9, 13, 17, 21], and achieve results in a range of 42.”
A Normalized Gaussian Wasserstein Distance for Tiny Object Detection (2021, unknown): “In addition, Optimal Transport Assignment (OTA) [6] formulates the label assignment process as an Optimal Transport problem from a global perspective.”