Walruspecs
Client-side object detection app deployed as a Walrus site. The app's dependencies are packaged into the project and published as a Walrus blob. The object detection model runs on the client's browser (thanks to transformers.js) without needing a web server for computation.
About
The project demonstrates the feasibility of deploying a packaged AI model to a Walrus blob, which would charge its users based on their usage. e.g. every time someone runs the object detection model, the user will pay the creator a fixed amount of money. The application has no external dependencies, which makes the app safe from other points of failures.
How it's made
This project uses transformers.js to enable an AI model run on the client side. For object detection, it uses detr-resnet-50 model with the confidence threshold of 50% i.e. any predictions with a confidence level below 50% will be omitted in the results. The application has been uploaded with Bootstrap's minified CSS/JS, transformers.js, and detr-resnet-50 ONNX model weights.