Hebbian Mirror:

Low level, multi threaded, http API service for image recognition. Written in Rust with CUDA support.

Dependencies:

This software requires:

• Rust nightly: rustc 1.40.0-nightly, cargo 1.40.0-nightly, rustup 1.20.2

• Rocket

• tch

• openssl

• Docker

• and more (please see Cargo.toml)

• please if building for IoT use proper compilation settings or build on device.

• please compile with IS_CUDA = false first, or check if your hardware is capable of CUDA before compiling

Deep Learning assets:

• Download weights $ wget -c https://github.com/LaurentMazare/ocaml-torch/releases/download/v0.1-unstable/yolo-v3.ot to assets folder. or run assets/downloads.sh

• use nightly: rustup default nightly

Build for development:

• cargo: $ RUST_LOG=info ROCKET_ENV=development cargo run
• docker: docker-compose -f docker-compose.dev.yml up

Run in production

• cargo: $ RUST_LOG=error ROCKET_ENV=production cargo run
• docker: docker-compose up

Build for production

• $ cargo build --release then chmod +x ./target/release/hebbian_mirror then $ RUST_LOG=error ROCKET_ENV=production ./target/release/hebbian_mirror

Tests:

Python script: - python3 -m venv venv, then source venv/bin/activate, then pip install -r requirements.txt and run test python e2e_test.py

Authors and Contributors:

• Lenart Bartosz
• Claire Amalfitano
• Janusz Roll

License:

• MIT

Rule to follow for this project

• Have a good reason for developing this service

How AI (yolo.v3) works:

YOLO v3 (you only look once) architecture

Predictions map (boxes)

Loss function calculation

Knowledge resources:

Very good explanation from Ayoosh Kathuria

finito