Automated evaluation
Score every model output automatically with metrics tuned to each modality. The evaluation set keeps growing.
Producing 3D outputs with generative AI?
results in an interactive 3D viewer.
with your team at the exact angle you're looking at.
outputs with your team in one click.
Formats
Surface reconstructions in .obj and .glb, surface-sampled for fair comparison.
Dense or sparse clouds in .ply, scored point-to-point against a reference.
.splat, .spz, .ksplat, and .sog files, GPU-sorted and rendered in the browser.
Evaluate model outputs automatically, then review and share them as living 3D, not screenshots.
Score every model output automatically with metrics tuned to each modality. The evaluation set keeps growing.
Every evaluation run links straight to the 3D output it scored. Jump from a number to the model it came from.
Leave comments directly on the 3D scene, anchored to a point and a camera angle. Feedback your team can act on.
Drop a link in chat or a pull request: it unfurls as a 2D preview, and clicking opens the interactive viewer at the exact same camera angle.
PSNR, SSIM, geometry D1/D2, topology, and splat attribute deltas, with error heatmaps that point to the geometry behind a score.
Three steps from your first model output to your team's first comment.
Send model outputs from your training and inference pipeline, straight into Cerforma.
Cerforma scores each output automatically and links it to an interactive 3D view.
Comment in the viewer and share camera-synced links in chat or pull requests.
Cerforma is MIT-licensed and self-hostable. Run the whole platform on your own infrastructure, read the source, and extend the evaluations to fit your models. No vendor lock-in, no per-seat metering. Your model outputs stay yours.
Join the BetaDrop in your own file (.obj, .glb, .ply, .splat, .spz), rotate, inspect, and share the exact camera angle you're looking at. No signup, no install.
Try the Viewer