When a method is published, the implementation goes out with it. Use it, cite it, tell us where it breaks.
Controllable image augmentation on Stable Diffusion + ControlNet. Generate targeted additions to a training set instead of random transforms, and keep the semantics intact with edge, pose, segmentation and face-mesh control plus quality-aware filtering (FID, Mahalanobis, PTD top-k/p).
A quantum-native database engine for workloads that live between Hilbert space and production. Store quantum circuits, state vectors and shot histograms alongside relational metadata. Query superpositions with SQL extensions, simulate decoherence and noise channels, and dispatch variational loops to QPUs or GPU simulators from one transactional log. Built for the lab's own work on quantum-enhanced generative models.
A small library for measuring how well a dataset covers the distribution you deploy into, before you spend money labelling more of it.
Permissive by default. Commercial use is fine. If a release depends on a partner's data, the data stays out and the code ships without it.
Links go here once each repository is public under the lab's account.
Ask for early access