I build tools for
seeing difficult data.

fastforks

Published vision models. Same output, faster.

Load a stereo, depth, segmentation, detection, or feature model by its id and get the original network's output, faster. Every model has a plain PyTorch port checked against the authors' code, plus our own CUDA kernels and, on Apple silicon, Metal kernels. Across the measured models the median gain over portable PyTorch is 1.9×, with the best at 10×.

• Models 174 models across 8 tasks: stereo, depth, multi-view, SAM, DINOv2/v3 & more
• Parity accuracy checked against the original code on public benchmarks
• Licences code and weights licences per model; 142 usable commercially