This project demonstrates how to process a video stream in real-time using WebAssembly and apply a pre-trained food classification model to each frame of the video in order to determine if food is present in that frame, all by integrating WasmEdge into YoMo serverless.

Open-source projects that we used:

• Serverless stream processing framework YoMo
• Integrate with WasmEdge to introduce WebAssembly, interop TensorFlow Lite model
• A deep learning model found on TensorFlow Hub; make sure to download TFLite (aiy/vision/classifier/food_V1), which was created by Google

Advantages:

• ⚡️ Low-latency: Streaming data processing applications can now be done in far edge data centers thanks to YoMo's highly efficient network services
• 🔐 Security: WasmEdge runs the data processing function in a WebAssembly sandbox for isolation, safety, and hot deployment
• 🚀 High Performance: Compared with popular containers, such as Docker, WasmEdge can be up to 100x faster at startup and have a much smaller footprint
• 🎯 Edge Devices: As WasmEdge consumes much less resources than Docker, it is now possible to run data processing applications on edge devices

$ git clone https://github.com/yomorun/yomo-wasmedge-tensorflow.git

$ curl -fsSL "https://get.yomo.run" | sh
$ yomo version
YoMo CLI version: v1.13.0

Details about YoMo CLI installation can be found here.

wget -qO- https://raw.githubusercontent.com/WasmEdge/WasmEdge/master/utils/install.sh | bash -s -- -e all -p /usr/local

If you have any questions about installation, please refer to the official documentation. Currently, this project works on Linux machines only.

$ sudo apt-get update
$ sudo apt-get install -y ffmpeg libjpeg-dev libpng-dev

Write app.go to integrate WasmEdge-tensorflow:

• Get WasmEdge-go:

  $ cd flow
  $ go get -u github.com/second-state/WasmEdge-go/wasmedge

• download the pretrained model file lite-model_aiy_vision_classifier_food_V1_1.tflite，and store to rust_mobilenet_food/src directory:

  $ wget 'https://storage.googleapis.com/tfhub-lite-models/google/lite-model/aiy/vision/classifier/food_V1/1.tflite' -O ./rust_mobilenet_food/src/lite-model_aiy_vision_classifier_food_V1_1.tflite

• install rustc and cargo

  $ curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
  $ export PATH=$PATH:$HOME/.cargo/bin
  $ rustc --version

• install wasm32-wasi target

  $ rustup target add wasm32-wasi

• compile wasm file

  $ cd rust_mobilenet_food
  $ cargo build --release --target wasm32-wasi
  # The output WASM will be `target/wasm32-wasi/release/rust_mobilenet_food_lib.wasm`.
  $ cd ..

• copy the compiled wasm file

  $ cp rust_mobilenet_food/target/wasm32-wasi/release/rust_mobilenet_food_lib.wasm .

• optional: compile wasm file to AOT mode

  wasmedgec rust_mobilenet_food_lib.wasm rust_mobilenet_food_lib.so

$ yomo serve -c ./zipper/config.yaml

$ cd flow
$ go run -tags tensorflow,image app.go

Download this demo video: hot-dog.mp4, and store to source directory：

$ wget -P source 'https://github.com/yomorun/yomo-wasmedge-tensorflow/releases/download/v0.2.0/hot-dog.mp4'

then run：

$ go run ./source/main.go ./source/hot-dog.mp4