Web Neural Network API Performance Comparison
Presenter: Wanming Lin
Duration: 2 min
Compare the performance of WebNN API native implementation with the polyfill (WebGL/Wasm) implementation.
Video
Hello everyone, this is Wanming from Intel.
In this video, I will compare the performance of Web Neural Network API with two implementations.
One is WebNN Native based on OpenVINO with CPU and GPU backends the other one is WebNN Polyfill based on TensorFlow.js with WebAssembly and WebGL backends.
I will run 3 pop Machine Learning samples.
First one is image classification.
This is a side-by-side comparison.
The left side is running on Chrome and the right side is running on Electron.js.
In this sample, I will run 3 models MobileNet, SqueezeNet, and ResNet.
I compared the WebNN Polyfill Wasm backend with OpenVINO CPU backend and WebNN Polyfill WebGL backend with OpenVINO GPU backend.
Look at the FPS panel, the WebNN Native is much ahead of the WebNN Polyfill.
Then I will run semantic segmentation sample using DeepLab model.
Last one is object detection.
In this sample I will run 2 models, TinYolo and SSD MobileNet.
This chart showcases very impressive performance gained by WebNN with native Machine Learning capabilities.
Scan this QR code to experience more wonderful WebNN samples.
Thank you very much for watching.