person_reidentification

Person Re-Identification

This repository contains training and inference code for person re-identification neural networks. The networks are based on the OSNet architecture provided by the torchreid project. The code supports conversion to the ONNX* format and inference of OpenVINO™ models.

Setup

Prerequisites

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

Installation

1. Create and activate virtual Python environment

cd $(git rev-parse --show-toplevel)/pytorch_toolkit/person_reidentification
bash init_venv.sh
. venv/bin/activate

Datasets

Networks were trained on the following datasets:

• Market-1501
• MSMT17v2

For training, set up a root directory to the datasets. Structure of the root directory is as follows:

root
├── market1501
│   └── Market-1501-v15.09.15
│       ├── bounding_box_test
│       ├── bounding_box_train
│       └── query
│
└── msmt17
    └── MSMT17_V2
        ├── mask_test_v2
        ├── mask_train_v2
        ├── list_gallery.txt
        ├── list_query.txt
        ├── list_train.txt
        └── list_val.txt

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: person-reidentification-retail-0300.yaml

Training

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

Example:

python main.py \
    --root /path/to/datasets/directory/root \
    --config config/person-reidentification-retail-0300.yaml

Test

To test your network, set in a configuration file the test.evaluate parameter to True and run a command like the one used for training. For visualization results, set the test.visrank parameter to True (it works only when test.evaluate is True). For visualization activation maps, set the test.visactmap parameter to True.

Pretrained Models

You can download pretrained models in the PyTorch format corresponding to the provided configs from a fileshare as well:

• person-reidentification-retail-0248
• person-reidentification-retail-0249
• person-reidentification-retail-0300

Test OpenVINO™ Re-Identification Models

OpenVINO™ models are represented by *.xml and *.bin files (Intermediate Representation (IR) format). To use such a model, set the next parameters in the config file:

model:
  openvino:
    name: /path/to/model/in/IR/format.xml
    cpu_extension: /path/to/cpu/extension/lib.so

NOTE: *.xml and *.bin files should be saved in the same directory.

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 convert_to_onnx.py \
       --config /path/to/config/file.yaml \
       --output-name /path/to/output/model \
       --verbose

   Name of the output model ends with .onnx automatically. By default, the output model path is model.onnx. Be careful about the load_weights parameter in the configurations file. The verbose argument is non-required and switches on detailed output in conversion function.

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  \
    --mean_values '[123.675, 116.28, 103.53]' \
    --scale_values '[58.395, 57.12, 57.375]' \
    --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-person tracking demo, which is able to use these models as person re-identification networks. See details in the demo.

Citation

If this training code helped your research, please cite the paper:

@misc{sovrasov2020building,
    title={Building Computationally Efficient and Well-Generalizing Person Re-Identification Models with Metric Learning},
    author={Vladislav Sovrasov and Dmitry Sidnev},
    year={2020},
    eprint={2003.07618},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}