Building desktop-class tools inside a web browser sandbox is incredibly challenging. In this post, we discuss the optimization layer that enables Fyloxa to run background removal algorithms and heavy image compression without uploading data.
GPU-Accelerated AI Background Removal
Typically, background removal requires sending an image to an API running on a costly server GPU. Fyloxa processes this via WebGL-accelerated ONNX runtimes. The neural network model runs directly inside your browser tab, leveraging your local device's graphics processor (GPU) to detect backgrounds in real time.
Efficient memory management
Processing files in-memory requires strict heap monitoring. We construct temporary object URLs using URL.createObjectURL to prevent browser tab crashes on heavy document sets, ensuring that data is automatically garbage collected when it is no longer required.