This is a pytorch implementation of the triplet loss. It is based on the blog post by Olivier Moindrot and the Facenet paper by Florian Schroff, Dmitry Kalenichenko and James Philbin.
Structure of the project
.github/
workflow/
tripletloss.yaml
experiments/
base_model/
params.json
param_search/
params.json
model/
__init__.py
data_loader.py
net.py
triplet_loss.py
tests/
__init__.py
test_triplet_loss.py
.gitignore
evaluate.py
LICENSE
Makefile
README.md
requirements.txt
search_hyperparams.py
synthesize_results.py
tox.ini
train.py
utils.py
visualization.py
The simplest way to use this repository as a template for a project is to clone it and then delete the .git directory. Then git can be re-initialized,
git clone <url> <newprojname>
cd <newprojname>
To start training we can do the following,
python train.py --data_dir=<wherever your dataset is>
To visualize the trained model, we can do the following,
python visualization.py
tensorboard --logdir=experiments/
I used Anaconda with python3,
conda create -n <yourenvname> python=<3.x>
conda activate <yourenvname>
conda install -n <yourenvname> --file requirements.txt