moyu284 / RLCC

RLCC Re-ID Baseline

This is the RLCC Re-id baseline. The codebase is modified from OpenUnReID project.

Installation

Requirements

• ubuntu 16.04+
• Python 3.5+
• PyTorch 1.1 or higher
• CUDA 9.0 or higher

We have tested the following versions of OS and softwares:

• OS: Ubuntu 16.04
• Python: 3.6/3.7
• PyTorch: 1.1/1.5/1.6
• CUDA: 9.0/11.0

Install OpenUnReID

a. Create a conda virtual environment and activate it.

conda create -n open-mmlab python=3.7 -y
conda activate open-mmlab

b. Install PyTorch and torchvision following the official instructions, e.g.,

conda install pytorch torchvision -c pytorch

c. Install the dependent libraries.

cd RLCC_Baseline
pip install -r requirements.txt

d. Install openunreid library.

python setup.py develop

e. Support AutoAugment. (optional)

You may meet the following error when using DATA.TRAIN.is_autoaug=True in config files,

AttributeError: Can't pickle local object 'SubPolicy.init..'

To solve it, you need to replace multiprocessing with multiprocess in torch.multiprocessing (generally found in $CONDA/envs/open-mmlab/lib/python3.7/site-packages/torch/multiprocessing/), e.g.

# refer to https://github.com/DeepVoltaire/AutoAugment/issues/16
import multiprocess as multiprocessing
from multiprocess import *

Prepare datasets

It is recommended to symlink your dataset root to OpenUnReID/datasets. If your folder structure is different, you may need to change the corresponding paths (namely DATA_ROOT) in config files.

Download the datasets from

• DukeMTMC-reID: [Google Drive]
• Market-1501-v15.09.15: [Google Drive]
• MSMT17_V1: [Home Page] (request link by email the holder)
• VehicleID_V1.0: [Home Page] (request link by email the holder)
• AIC20_ReID_Simulation (VehicleX): [Home Page] (request password by email the holder)
• VeRi_with_plate: [Home Page] (request link by email the holder)

Save them under

OpenUnReID
└── datasets
    ├── dukemtmcreid
    │   └── DukeMTMC-reID
    ├── market1501
    │   └── Market-1501-v15.09.15
    ├── msmt17
    │   └── MSMT17_V1
    ├── personx
    │   └── subset1
    ├── vehicleid
    │   └── VehicleID_V1.0
    ├── vehiclex
    │   └── AIC20_ReID_Simulation
    └── veri
        └── VeRi_with_plate

Getting Started

The training and testing scripts can be found in OpenUnReID/tools. We use 4 GPUs for training and testing, which is considered as a default setting in the scripts. You can adjust it (e.g. ${GPUS}, ${GPUS_PER_NODE}) based on your own needs.

Test

Testing commands

• Distributed testing with multiple GPUs:

bash tools/dist_test.sh ${RESUME}

• Testing with a single GPU:

GPUS=1 bash tools/dist_test.sh ${RESUME}

Arguments

• ${RESUME}: model for testing, e.g. ../logs/20210101/market1501/model_best.pth.

Configs

• Test with different datasets, e.g.

TEST:
  datasets: ['market1501',] # arrange the names in a list

• Add re-ranking post-processing, e.g.

TEST:
  rerank: True # default: False

• Save GPU memory but with a lower speed,

TEST:
  dist_cuda: False # use CPU for computing distances, default: True
  search_type: 3 # use CPU for re-ranking, default: 0 (1/2 is also for GPU)

Train

Training commands

• Training with single node multiple GPUs:

CUDA_VISIBLE_DEVICES=0,1,2,3 bash tools/dist_train.sh ${WORK_DIR} 

Arguments

• ${WORK_DIR}: folder for saving logs and checkpoints, e.g. 20210101/market1501, the absolute path will be LOGS_ROOT/${WORK_DIR} (LOGS_ROOT is defined in config files).

Configs

• Flexible architectures,

MODEL:
  backbone: 'resnet50' # or 'resnet101', 'resnet50_ibn_a', etc
  pooling: 'gem' # or 'avg', 'max', etc
  dsbn: True # domain-specific BNs, critical for domain adaptation performance

• Ensure reproducibility (may cause a lower speed),

TRAIN:
  deterministic: True

• Dataset Config, the conventional USL task, e.g. unsupervised market1501

TRAIN:
  # arrange the names in a dict, {DATASET_NAME: DATASET_SPLIT}
  datasets: {'market1501': 'trainval'}
  # val_set of 'market1501' will be used for validation
  val_dataset: 'market1501'

• Mixed precision training

TRAIN:
  amp: True # mixed precision training for PyTorch>=1.6