vehicle_reidentification

Vehicle Re-Identification

This repository contains training and inference code for vehicle re-identification neural networks. The networks are based on the OSNet architecture provided by the deep-object-reid project. The code supports conversion to the ONNX* format.

Model Name	VeRi-776* rank-1	VeRi-776* mAP	GFlops	MParams	Links
vehicle-reid-0001	96.31	85.15	2.643	2.183	shapshot, configuration file

Setup

Prerequisites

• Ubuntu* 16.04
• Python* 3.5.2
• PyTorch* 1.3 or higher
• OpenVINO™ 2019 R4 (or newer) with Python API

Installation

To create and activate virtual Python environment follow installation instructions

Datasets

This toolkit contains configs for training on the following datasets:

• VeRi-776
• VeRi-Wild
• UniverseModels (set of make/model classification datasets with merged annotation)
  • CompCars
  • VMMRdb

Note: The instruction how to prepare the training dataset can be found in DATA.md

The final Structure of the root directory is as follows:

root
├── veri
│   ├── image_train
│   ├── image_query
│   ├── image_test
│   └── train_label.xml
│
├── veri-wild
│   ├── images
│   └── vehicle_info.txt
│
└── universe_models
    └── images

Configuration Files

The script for training and inference uses a configuration file default_config.py, which consists of default parameters. This file also has description of parameters. Parameters that you wish to change must be in your own configuration file. Example: vehicle-reid-0001.yaml

Training

To start training, create or choose a configuration file and use the main.py script.

Example:

python ../../../external/deep-object-reid/tools/main.py \
    --root /path/to/datasets/directory/root \
    --config configs/vehicle-reid-0001.yaml

Fine-tuning

To start fine-tuning, create or choose a configuration file, choose init model weigts (you can use the pre-trained weights -- see section) and use the main.py script.

Example:

python ../../../external/deep-object-reid/scripts/main.py \
    --root /path/to/datasets/directory/root \
    --config configs/vehicle-reid-0001.yaml \
    model.load_weights /path/to/pretrained/model/weigts

Testing

To test your network, specify your configuration file and use the main.py script.

Example:

python ../../../external/deep-object-reid/scripts/main.py \
    --root /path/to/datasets/directory/root \
    --config configs/vehicle-reid-0001.yaml \
    model.load_weights /path/to/trained/model/weigts \
    test.evaluate True

Convert a PyTorch Model to the OpenVINO™ Format

Follow the steps below:

1. Convert a PyTorch model to the ONNX format by running the following:

   python ../../../external/deep-object-reid/tools/convert_to_onnx.py \
       --config /path/to/config/file.yaml \
       --output-name /path/to/output/model \
       model.load_weights /path/to/trained/model/weigts

   Name of the output model ends with .onnx automatically. By default, the output model path is model.onnx.

2. Convert the obtained ONNX model to the IR format by running the following command:

python <OpenVINO_INSTALL_DIR>/deployment_tools/model_optimizer/mo.py --input_model model.onnx  \
    --input_shape [1,3,208,208]
    --reverse_input_channels

This produces the model.xml model and weights model.bin in single-precision floating-point format (FP32).

OpenVINO™ Demo

OpenVINO™ provides the multi-camera-multi-target tracking demo, which is able to use these models as vehicle re-identification networks. See details in the demo.

Citation

Original repository: github.com/sovrasov/deep-person-reid/tree/vehicle_reid