pymetric tries to build a metric learning and retrieval codebase based on PyTorch. It refers from pycls On Network Design Spaces for Visual Recognition project and fast-reid from (https://github.com/JDAI-CV/fast-reid).

• 2nd place in Google Landmark Retrieval 2020 Competition. (https://drive.google.com/file/d/1XnzxMOHhzua9tjrAjo-X55ieKVJzRJw_/view?usp=sharing)
• Includes pycls (https://github.com/facebookresearch/pycls), refer to pycls.md.
• pymetric is written by DistributedDataParallel which is different from fast-reid. Now mainly includes features such as arcface loss and circle loss, ongoing.
• Add multicards feature extraction, searching topk and computing mAP.

Requirements:

• NVIDIA GPU, Linux, Python3(tested on 3.6.10)
• PyTorch, various Python packages; Instructions for installing these dependencies are found below

Notes:

• pymetric does not currently support running on CPU; a GPU system is required
• pymetric has been tested with CUDA 10.2 and cuDNN 7.1

To install PyTorch with CUDA support, follow the installation instructions from the PyTorch website.

Clone the pymetric repository:

# PYMETRIC=/path/to/clone/pymetric
git clone https://github.com/feymanpriv/pymetric $PYMETRIC

Install Python dependencies:

pip install -r $PYMETRIC/requirements.txt

Set PYTHONPATH:

cd $PYMETRIC && export PYTHONPATH=`pwd`:$PYTHONPATH

Same with pycls, pymetric finds datasets via symlinks from metric/datasets/data to the actual locations where the dataset images and annotations are stored. Refer to DATA.md.

Training a metric model:

python tools/train_metric.py \
    --cfg configs/metric/R-50-1x64d_step_8gpu.yaml \
    OUT_DIR ./output \
    PORT 12001 \
    TRAIN.WEIGHTS path/to/pretrainedmodel

Resume training:

python tools/train_metric.py \
    --cfg configs/metric/R-50-1x64d_step_8gpu.yaml \
    OUT_DIR ./output \
    PORT 12001 \
    TRAIN.AUTO_RESUME True

Extracting features(labels) and evaluation

set ${total_num} = n*(gpu_cards)
sh tools/metric/eval/infer.sh
python search.py search_gpu ${queryfea_path}, ${referfea_path}, ${output}

Convert to tensorflow2.3 (please refer onnx and onnx-tensorflow)
examples: tools/convert/torch2onnx.py tools/convert/onnx2tf.py

-resnet50 (c3l3)

-resnet101 (t3ln)

-resnest269 (3c5a)

2nd place on Google Landmark Retrieval Challenge 2020

pymetric is released under the MIT license. Please see the LICENSE file for more information.