ALL CONTENT IS DIRECTLY FROM THE GITHUB README

RF-DETR: SOTA Real-Time Object Detection Model

• hf space link
• colab demo link
• roboflow blog link
• discord link

RF-DETR is a real-time, transformer-based object detection model architecture developed by Roboflow and released under the Apache 2.0 license.

RF-DETR is the first real-time model to exceed 60 AP on the Microsoft COCO benchmark alongside competitive performance at base sizes. It also achieves state-of-the-art performance on RF100-VL, an object detection benchmark that measures model domain adaptability to real world problems. RF-DETR is comparable speed to current real-time objection models.

RF-DETR is small enough to run on the edge, making it an ideal model for deployments that need both strong accuracy and real-time performance.

Results

We validated the performance of RF-DETR on both Microsoft COCO and the RF100-VL benchmarks.

https://media.roboflow.com/rf-detr/charts.png" alt="rf-detr-coco-rf100-vl-8">

| Model | mAP

Training parameters of the improved DETR network model.

This study presents the development and application of an optimized Detection Transformer (DETR) model, known as CD-DETR, for the detection of thoracic diseases from chest X-ray (CXR) images. The CD-DETR model addresses the challenges of detecting minor pathologies in CXRs, particularly in regions with uneven medical resource distribution. In the central and western regions of China, due to a shortage of radiologists, CXRs from township hospitals are concentrated in central hospitals for diagnosis. This requires processing a large number of CXRs in a short period of time to obtain results. The model integrates a multi-scale feature fusion approach, leveraging Efficient Channel Attention (ECA-Net) and Spatial Attention Upsampling (SAU) to enhance feature representation and improve detection accuracy. It also introduces a dedicated Chest Diseases Intersection over Union (CDIoU) loss function to optimize the detection of small targets and reduce class imbalance. Experimental results on the NIH Chest X-ray dataset demonstrate that CD-DETR achieves a precision of 88.3% and recall of 86.6%, outperforming other DETR variants by an average of 5% and CNN-based models like YOLOv7 by 6–8% in these metrics, showing its potential for practical application in medical imaging diagnostics.

mAP of the model before and after data augmentation.

Single-frame image inference time of different detection methods.

Model and parameters of each device in the acquisition device.

Numbers of pinhole defects before and after data augmentation.

Metallized Ceramic Ring is a novel electronic apparatus widely applied in communication, new energy, aerospace and other fields. Due to its complicated technique, there would be inevitably various defects on its surface; among which, the tiny pinhole defects with complex texture are the most difficult to detect, and there is no reliable method of automatic detection. This Paper proposes a method of detecting micro-pinhole defects on surface of metallized ceramic ring combining Improved Detection Transformer (DETR) Network with morphological operations, utilizing two modules, namely, deep learning-based and morphology-based pinhole defect detection to detect the pinholes, and finally combining the detection results of such two modules, so as to obtain a more accurate result. In order to improve the detection performance of DETR Network in aforesaid module of deep learning, EfficientNet-B2 is used to improve ResNet-50 of standard DETR network, the parameter-free attention mechanism (SimAM) 3-D weight attention mechanism is used to improve Sequeeze-and-Excitation (SE) attention mechanism in EfficientNet-B2 network, and linear combination loss function of Smooth L1 and Complete Intersection over Union (CIoU) is used to improve regressive loss function of training network. The experiment indicates that the recall and the precision of the proposed method are 83.5% and 86.0% respectively, much better than current mainstream methods of micro defect detection, meeting requirements of detection at industrial site.

Settings of different models and comparison of detection results.

Comparison of performances of different detection methods.