layumi/Person_reID_baseline_pytorch

:bouncing_ball_person: Pytorch ReID: A tiny, friendly, strong pytorch implement of person re-id / vehicle re-id baseline. Tutorial 👉https://github.com/layumi/Person_reID_baseline_pytorch/tree/master/tutorial

What it solves

This project provides a strong and accessible baseline for Object Re-identification (re-ID), specifically focused on person re-identification. It allows users to identify and track the same object (person) across different non-overlapping camera views, solving the problem of consistent object retrieval in surveillance or security contexts.

How it works

The project is built on PyTorch and implements a variety of deep learning architectures and loss functions to learn discriminative features for object retrieval. It supports multiple backbones such as ResNet, DenseNet, HRNet, Swin Transformer, and DinoV3. To improve accuracy, it incorporates advanced training techniques like Circle Loss, Triplet Loss, and Random Erasing, as well as post-processing methods like Re-Ranking to refine retrieval results.

Who it’s for

This is designed for researchers and developers in computer vision who need a reliable, lightweight baseline for re-ID tasks, as well as beginners who are new to object re-identification and can leverage the included tutorials.

Highlights

  • Low Hardware Requirements: Supports bf16/fp16 precision, allowing training on GPUs with as little as 2GB of memory.
  • Extensive Model Support: Includes a variety of architectures from traditional CNNs (ResNet, DenseNet) to modern Transformers (Swin, DinoV3).
  • Comprehensive Loss Functions: Implements multiple metric learning losses including Circle, Triplet, Contrastive, Sphere, and Arcface.
  • Developer Friendly: Provides a dedicated tutorial for beginners and off-the-shelf options to apply state-of-the-art tricks in a single line of code.
  • Performance Optimized: Includes GPU-accelerated re-ranking and support for TensorRT and PyTorch JIT for faster inference.

Related

  • Project
    pytorch/benchmark

    PyTorch Benchmarks is a collection of standardized, lightweight versions of popular deep‑learning models (BERT, ResNet, Stable Diffusion, etc.) that can be run to measure the performance of different PyTorch builds, CUDA versions, and back‑ends. It offers a uniform API, install scripts, and multiple runners (simple test, pytest‑benchmark, custom userbenchmark) plus utilities for low‑noise tuning on AWS g4dn.metal instances.

  • Project
    LucasAlegre/morl-baselines

    MORL‑Baselines is a PyTorch library of reliable multi‑objective reinforcement‑learning algorithms (single‑ and multi‑policy, SER/ESR). It follows the MO‑Gymnasium API, provides utilities, automatic W&B logging, and integrates with the Open RL Benchmark for reproducible experiments.

  • Project
    pyg-team/pytorch_geometric

    PyTorch Geometric (PyG) is a PyTorch‑based library for building, training, and scaling graph neural networks. It offers a unified API, dozens of ready‑made GNN layers, data loaders for large and heterogeneous graphs, and easy extensibility via a `MessagePassing` base class.

  • Dispatch
    Profiling in PyTorch: A Beginner's Guide to torch.profiler

    Hugging Face provides a comprehensive guide to using torch.profiler to identify bottlenecks, understand the CPU-GPU dispatch chain, and analyze the impact of torch.compile on kernel execution.

  • Project
    facebookresearch/recipes

    A library of ready-to-run deep learning training recipes that provide production-ready baselines for common tasks to eliminate repetitive glue code.