Tianxiaomo/pytorch-YOLOv4
Pytorch YOLOv4
This project is a PyTorch implementation of the YOLOv4 object detection framework. It provides a system for training and deploying neural networks that identify and locate multiple objects within images and video streams.
The framework includes tools for converting trained weights into universal formats and hardware-specific optimized engines, specifically supporting ONNX and TensorRT. It features a TensorRT inference optimizer to reduce latency and increase throughput, as well as a model architecture compatible with NVIDIA DeepStream streaming analytics pipelines.
The system covers model training with mosaic data augmentation and supports real-time object detection across multiple inference engines. It provides utilities for model conversion to TensorFlow and supports both static and dynamic batch configurations for deployment.
Features
- Object Detection - Implements a full object detection system to identify and locate multiple objects within images and video streams.
- Real-Time Object Detection - Implements a real-time object detection framework capable of processing live video streams across multiple inference engines.
- Detection Model Training - Provides a complete system for training object detection models on custom datasets using YOLO format configurations.
- Inference Model Deployment - Implements the conversion of trained models into optimized compute graphs for hardware-specific execution via ONNX and TensorRT.
- TensorRT Engine Compilation - Includes a TensorRT inference optimizer to compile weights into high-performance engines for reduced latency.
- YOLOv4 Implementations - Provides a complete PyTorch implementation of the YOLOv4 object detection architecture.
- Mosaic Augmentations - Implements mosaic data augmentation by combining four training images to improve detection across various scales.
- Deep Learning Model Converters - Includes tools to translate trained neural network weights into ONNX and TensorRT formats for deployment.
- DeepStream Compatible Models - Offers a detection architecture specifically configured for seamless integration into real-time DeepStream streaming pipelines.
- DeepStream Pipeline Integrations - Provides a model architecture specifically configured for integration into NVIDIA DeepStream streaming analytics pipelines.
- Streaming Analytics Integrations - Provides a model architecture compatible with NVIDIA DeepStream streaming analytics pipelines for scalable video processing.
- CSPDarknet Architectures - Implements a CSPDarknet-based backbone to reduce computational overhead during visual feature extraction.
- Image Inference Clients - Provides an inference system that accepts images of flexible dimensions and returns detected objects using various weight formats.
- Inference Performance Optimization - Optimizes model execution speed and reduces latency by compiling weights into specialized TensorRT engines.
- ONNX Model Exporters - Provides utilities to convert trained PyTorch weights into the standardized ONNX format for cross-platform deployment.
- ONNX-to-TensorRT Conversions - Converts exported models into optimized TensorRT engines to increase inference speed on NVIDIA hardware.
- TensorRT Engine Generators - Ships an engine that compiles model weights into high-performance TensorRT runtime engines for NVIDIA GPUs.
- Inference Pipelines - Implements an end-to-end inference pipeline supporting interchangeable backends like PyTorch, ONNX Runtime, and TensorRT.
- Mosaic Augmentations - Implements mosaic data augmentation to improve the model's ability to detect objects across different scales.
- Computer Vision Libraries - YOLOv4 implementation for PyTorch and TensorRT.
- CNN - Listed in the “CNN” section of the Ailia Models awesome list.
Star history
How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.
