D

Deta Swin Large

Developed by jozhang97
DETA is a transformer-based object detection model that achieves rapid convergence and efficient detection by reintroducing the IoU assignment mechanism and NMS methods.
Downloads 2,741
Release Time : 1/30/2023

Model Overview

DETA reintroduces Intersection over Union (IoU) assignment mechanism and Non-Maximum Suppression (NMS) methods for transformer-based detectors, significantly improving training efficiency and detection performance.

Model Features

Fast convergence
Achieves 50.2 mAP on the COCO dataset in just 12 epochs, with significantly faster convergence than similar models.
Efficient training
Training and testing speeds are comparable to Deformable DETR, but with superior performance.
Improved assignment mechanism
Reintroduces IoU assignment mechanism and NMS methods to enhance detection accuracy.

Model Capabilities

Object detection
Image analysis

Use Cases

Computer vision
General object detection
Detects and locates multiple objects in complex scenes
Achieves 50.2 mAP on the COCO dataset
Real-time surveillance systems
Used for object detection and tracking in video surveillance
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase