FIREBAL runs Richardson–Lucy deconvolution on the GPU to reverse optical blur in confocal, widefield and light-sheet stacks — no CUDA setup, no desktop install. Upload a stack, generate or supply a PSF, and get a restored volume back.
Traditional deconvolution means wrangling a local CUDA toolchain, matching driver versions, and hand-tuning tiling to fit your GPU. FIREBAL moves all of that behind a web UI backed by serverless GPU workers.
An iterative RL deconvolution engine written in CUDA C++, with regularisation and Wiener/OTF controls exposed per job.
Synthesise widefield, confocal and light-sheet point-spread functions from optical parameters — or upload your own measured PSF.
Large volumes are split into VRAM-sized tiles automatically, with overlap blending so you never see seams in the result.
Multi-channel, multi-timepoint stacks are handled natively and returned as ImageJ-compatible TIFFs.
Jobs run on on-demand GPU workers and scale to zero when idle — you get the compute without renting a machine.
The entire workflow lives in the browser. Open the app, drop in a stack, and download the restored volume.
Drag in a TIFF stack. Large files stream straight to object storage, bypassing upload size limits.
Set the axis order, pick or generate a PSF, and tune iterations and regularisation.
A GPU worker tiles the volume, runs Richardson–Lucy, and reassembles the result.
Pull back an ImageJ-ready TIFF — sharper, with the optical blur reversed.
Create a verified account to process images. Self-registered accounts begin on the free tier.
FIREBAL is built by a small group working at the intersection of microscopy, GPU computing and scientific tooling.
Designs and maintains the Richardson–Lucy engine and the GPU processing pipeline.
Builds the web app, the serverless GPU deployment and the object-storage backend.
Researchers contributing sample data, optical parameters and validation across imaging modalities.
Open the app and run your first deconvolution in the browser.