Face technology

Detection, landmark localisation, face parsing and attribute estimation, evaluated under the conditions mobile capture actually produces: off-angle pose, low illumination and motion.

Portrait retouching

Correction of skin tone, blemishes and specular highlights while preserving the fine texture that distinguishes a photograph of a face from a rendering of one.

Image restoration

Denoising, deblurring, super-resolution and artefact repair, applied to single frames and to sequences, where degradation originates in the capture rather than the encoding.

Segmentation

Separation of subject, hair, sky and background at a boundary accuracy that withstands close inspection, with temporal stability sufficient to avoid mask flicker in motion.

Relighting & colour

Estimation and adjustment of exposure, white balance and light direction such that the edited region remains photometrically consistent with the rest of the frame.

Image generation

Synthesis of backgrounds, inpainted regions and canvas extensions conditioned on the surrounding frame, matching its grain, optical characteristics and illumination.

On-device inference

Measurement on target devices rather than on reference hardware. For video this imposes a per-frame budget, making latency, memory footprint and power draw part of the reported result.

Temporal consistency

Maintaining an identical edit across adjacent frames, the constraint that distinguishes video processing from the frame-by-frame application of a still-image method.