/ossmi

A Simple MTCNN Face Detection App

Primary LanguagePython

Ossmi. An Interactive Face Search App

Standalone (web or mobile) personal photo gallery with face recognition. Google Photos на минималках?

  • Dynamically indexing faces from newly uploaded photos
  • Searching photos of a person within a gallery by leveraging indexed faces
  • Annotating/naming persons (face_ids)

demo

System Design

Todo:

  • Dynamically update indexes with new images

  • Automate the creation of new collections and indexes

  • More data: collect more images from kipyatcom, ppz, vk

  • Consider using VGGFace2 for demo?

  • Fine tune the Inceptionesnet v1 on a new dataset

  • Analyze model performance, choose optimal distance metric (Euclidean and Cosine)

  • Tune and evaluate faiss indexing

  • Data Visualization. Analyze outliers

  • Data Augmentation

  • Publish on Heroku, AWS, GCP?

  • Testing

  • Documentation

  • Write a blog on how to use the app?

  • Publish on ProductHunt?

Bugs

  • Faces on a rotated or tilted images are not detected

Usage

  1. Install streamlit: python3 -m pip install streamlit

  2. Run the app: streamlit run demo/demo.py

Minio

docker run -d -p 9000:9000 -e MINIO_ACCESS_KEY=minioadmin -e MINIO_SECRET_KEY=minioadmin -v /mnt/data:/data minio/minio server /data

PostgreSQL

docker run --name ossmi -e POSTGRES_PASSWORD=postgres -d -p 15432:5432 -v pgdata:/var/lib/postgresql/data postgres

Milvus Docker Reference

Connecting to aws ec2 instance

ssh -i mtcnn_streamlit.pem ubuntu@ec2-35-155-200-213.us-west-2.compute.amazonaws.com