facenet

KO | EN

gitlite — search

#python #hacktoberfest #react #ai #android #machine-learning #deep-learning #macos #windows #cli #kubernetes #pytorch #mcp #openai

(★ 251)

FaceNet for face recognition using pytorch

파일 탐색기

최종 버전 다운로드 (.zip)

- __init__.py
- write_csv_for_making_dataset.py
- - accuracy.jpg
  - loss.jpg
  - roc.png
  - test-result.png
  - threshold.png
- .gitignore
- train.csv
- valid.csv
.gitignore
data_loader.py
eval_metrics.py
hubconf.py
lfw.ipynb
LICENSE
loss.py
models.py
mtcnn.py
plotter.ipynb
README.md
test-result.ipynb
threscounts.csv
train.py
utils.py

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Facenet for face verification using pytorch
- Pytorch implementation of the paper: "FaceNet: A Unified Embedding for Face Recognition and Clustering".
- Training of network is done using triplet loss.
- This work is modified in some functionality from the original work by [Taebong Moon](https://github.com/tbmoon/facenet) and then retrained for the purpose of completing my BS degree. The full report can be found at this folder: [full-and-paper-report](https://github.com/khrlimam/skripsi/raw/master/pdfs)
- To use the pretrained model please refer to this repo: https://github.com/khrlimam/res-facenet
- If you wish to try the demo app please clone this repo and follow the installation instruction: https://github.com/khrlimam/demo-facenet

# How to train/validate model
- Download vggface2 (for training) and lfw (for validation) datasets.
  - https://www.robots.ox.ac.uk/%7Evgg/data/vgg_face2/
  - http://vis-www.cs.umass.edu/lfw/lfw.tgz

- Align face image files by following David Sandberg's instruction (part of "Face alignment").
  - https://github.com/davidsandberg/facenet/wiki/Classifier-training-of-inception-resnet-v1

- Write list file of face images by running "datasets/write_csv_for_making_dataset.py"
  - `python write_csv_for_making_dataset.py --root-dir=/path/to/dataset/dir --final-file=dataset.csv`
  - `datasets/write_csv_for_making_dataset.py` is multiprocess version of previous `.ipynb`. This way generating csv dataset is much faster.
  
- Train
  - Again, one need to modify paths in accordance with location of image dataset.
  - Also feel free to change some parameters.
  
  
# Results

- Accuracy on VGGFace2 and LFW datasets

![accuracy](https://raw.githubusercontent.com/tbmoon/facenet/master/log/tmp/accuracy.jpg)

- Triplet loss on VGGFace2 and LFW datasets

![loss](https://raw.githubusercontent.com/tbmoon/facenet/master/log/tmp/loss.jpg)

- ROC curve on LFW datasets for validation

![roc curve](https://raw.githubusercontent.com/tbmoon/facenet/master/log/tmp/roc.png)

- True counts on each threshold

![True counts on each threshold](https://raw.githubusercontent.com/tbmoon/facenet/master/log/tmp/threshold.png)

- Test the model on 30 pair of images with threshold 1.5
![Test Result on 30 pair of images](https://raw.githubusercontent.com/tbmoon/facenet/master/log/tmp/test-result.png)

# References
- https://github.com/liorshk/facenet_pytorch 
- https://github.com/davidsandberg/facenet
- https://arxiv.org/abs/1503.03832
- https://arxiv.org/abs/1703.07737

facenet

파일 탐색기

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

facenet

파일 탐색기

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README