Dataset

This dataset was created by Nikita Sherstnev

Contents

RF DETR Waste

## Overview

RF DETR Waste is a dataset for object detection tasks - it contains Waste Recycling annotations for 2,958 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [Public Domain license](https://creativecommons.org/licenses/Public Domain).

This dataset is intended for the reproducability of the methods in 'Exploration of TPU Architectures for the Optimized Transformer in Drainage Crossing Detection', along with the provided GitHub repository at https://github.com/SHUs-Lab/BTSD24AN.The dataset consists of 6,012 LiDAR-derived DEM georeferenced rasters in TIF format, each with an 800m x 800m extent and a cell size of 1m x 1m. Elevation data is stored as 32-bit floating point values, indicating meters above sea level, and comes from the USGS 3DEP program.The rasters cover four watersheds in the Continental United States: Sacramento-Stone Corral in California, Vermilion River in Illinois, Maple River in North Dakota, and West Fork Big Blue in Nebraska. Drainage crossings within these watersheds were labeled as centroids, and corresponding rasters containing these centroids were extracted. Bounding boxes of 100m x 100m were defined around these centroids, and the data were converted to the COCO format for use with the DETR model.After filtering out anomalous rasters, 6,007 rasters with 13,141 drainage crossing bounding boxes were used. The Maple River Watershed data was reserved for transfer learning.The directory structure is as follows:processed_data├── initial_data│ ├── annotations│ ├── test│ ├── train│ └── validate└── transfer_data ├── annotations └── test

The dataset is structured for person object detection tasks, containing separate directories for training, validation, and testing. Each split has an images folder with corresponding images and a labels folder with annotation files.

Train Set: Contains images and annotations for model training.

Validation Set: Includes images and labels for model evaluation during training.

Test Set: Provides unseen images and labels for final model performance assessment.

Each annotation file (TXT format) corresponds to an image and likely contains bounding box coordinates and class labels. This structure follows standard object detection dataset formats, ensuring easy integration with detection models like yolo,RT-DETR.

Dataset Structure

📂 dataset/ ├── 📁 train/ │ ├── 📂 images/ │ │ ├── 🖼 image1.jpg (Training image) │ │ ├── 🖼 image2.jpg (Training image) │ ├── 📂 labels/ │ │ ├── 📄 image1.txt (Annotation for image1.jpg) │ │ ├── 📄 image2.txt (Annotation for image2.jpg) │ ├── 📁 val/ │ ├── 📂 images/ │ │ ├── 🖼 image3.jpg (Validation image) │ │ ├── 🖼 image4.jpg (Validation image) │ ├── 📂 labels/ │ │ ├── 📄 image3.txt (Annotation for image3.jpg) │ │ ├── 📄 image4.txt (Annotation for image4.jpg) │ ├── 📁 test/ │ ├── 📂 images/ │ │ ├── 🖼 image5.jpg (Test image) │ │ ├── 🖼 image6.jpg (Test image) │ ├── 📂 labels/ │ │ ├── 📄 image5.txt (Annotation for image5.jpg) │ │ ├── 📄 image6.txt (Annotation for image6.jpg)

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.

The dataset (Lego_Tracking folder) has been created manually by recording 12 videos by smartphone. Of these, 10 were designated for training and 2 for testing. The videos showcase conveyor belts transporting LEGO bricks, captured from various perspectives (top, front, and diagonal) to provide diverse viewpoints. The videos have been recorded in the AI laboratory of the Eötvös Loránd University (https://github.com/BahruzHuseynov/Object-Tracking-AI_Lab) and the dataset has been used to make a research about detection-segmentation-tracking pipeline to complete the AI laboratory work. The dataset includes videos of differing complexities, classified as "Overlapping," "Normal," or "Simple," with varying durations ranging from short to long shots. Additionally, the annotation of LEGO bricks was performed frame-by-frame using the RoboFlow web application (https://roboflow.com/).

In addition, you can go to the dataset which has been prepared by applying systematic sampling on training videos to train and validate YOLOv8 and RT-DETR models from ultralytics: https://www.kaggle.com/datasets/hbahruz/multiple-lego/data

Another dataset prepared can be used for the semantic segmentation: https://www.kaggle.com/datasets/hbahruz/lego-semantic-segmentation/

Single-frame image inference time of different detection methods.

Model and parameters of each device in the acquisition device.

Settings of different models and comparison of detection results.

Comparison of performances of different detection methods.

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.