No description available

A fork of the official rf-detr repo with some changes to avoid stretching all images to be squares. Instead, we have all images keep their native aspect ratio and use padding to establish a uniform image size during training.

RF-DETR fine-tuning for drone detection - training pipeline with ONNX/TensorRT export for NVIDIA Jetson deployment

No description available

a guide on how to train an rf detr instance segmentation model

This project provides complete RF-DETR training and real-time ONNXRuntime inference

This project provides complete RF-DETR training and real-time ONNXRuntime inference

RF-DETR Training and Evaluation on BDD100K dataset as part of an Applied CV coding assignment by Bosch.

End-to-end RF-DETR segmentation workflow: YOLO-to-COCO conversion, 50-epoch training, and batch prediction export.

# RF-DETR Clean: Minimal DETR Training Pipeline (COCO-Style, Custom Dataset) This repository contains a lightweight DETR-style model (“DETRBaby”) and a clean training pipeline designed for custom object detection datasets using the COCO format. It has been optimized for two environments:

Ongoing project creating scripts to convert Google OpenImages to COCO format for Roboflow Detection Transformer (RF-DETR) training. Training object detection models on 60K+ images using H200 Research Cluster.

This project implements an end-to-end object detection system using the RF-DETR (Receptive Field Enhanced Detection Transformer) model on a custom image dataset. Built with PyTorch, the pipeline includes data preparation, model architecture, training, evaluation, and visualization.

This project trains and evaluates an RF-DETR object detection model on a custom COCO-formatted animal dataset from Roboflow. After 15 epochs of training with augmentation and gradient accumulation, performance is assessed using mAP, accuracy, precision, recall, F1 score, and visualizations via the Supervision library.