A training-free framework enabling interactive, layer-based editing using latent diffusion models. Our method achieves real-time edits (≈140 ms per edit on consumer GPUs) and was presented at ICCV 2025.
Demo teaser
Face tracking
From 2018 to 2020, I worked at Mirametrix as a research scientist, designing neural-network and computer-vision systems for face identification, accurate 2D/3D face and eyelid landmark estimation, 3D head-pose tracking, attention sensing, and point-of-gaze estimation.
OCT image Segmentation
My M.Sc. research focused on image processing and machine learning algorithms for biomedical image analysis, particularly Optical Coherence Tomography (OCT). The shape and thickness of retinal layers can change under different pathological conditions, so part of my work developed an efficient segmentation algorithm for delineating retinal layers and supporting ophthalmologists in diagnosis.
Bioprinting of Chronic wounds
In this project I designed and implemented an end-to-end tool for the bio-printing of chronic wounds.
Chronic wounds are still one of the major challenges of the dermatology science. Using the proposed method, skin cells can be easily printed on the corresponding coordinates using a bio-compatible solution and the wound healing procedure can be accelerated dramatically.
The tool delineates the accurate 3D geometry of wounds using image processing and computer vision techniques. Afterwards, a 3D model is constructed based on those coordinates and the bioprinter robot will complete the process. The results of this project are published in Journal of Biomedical and Health Informatics.
© Copyright 2026 Peyman Gholami