Projects
Research projects and open-source tools.
Research
RePaint
Image inpainting using Denoising Diffusion Probabilistic Models. Free-form inpainting with pre-trained unconditional DDPMs — no task-specific training needed.
SMILE
Semantically-guided Multi-attribute Image and Layout Editing. Joint manipulation of image content and spatial layout.
GANmut
Learning Interpretable Conditional Space for Gamut of Emotions. Continuous emotion space learning with disentangled control.
SESAME
Semantic Editing of Scenes by Adding, Manipulating or Erasing Objects. Scene-level editing with semantic control via GANs.
DeepSEE
Deep Disentangled Semantic Explorative Extreme Super-Resolution. Explores the space of plausible super-resolution outputs with semantic control.
SMIT
Stochastic Multi-Label Image-to-Image Translation. Diverse multi-domain translations with stochastic style codes and multi-label conditioning.
Independent
trajviz
Fast 3D camera trajectory video renderer using NumPy + OpenCV → ffmpeg. Visualize camera paths in 3D space. Published on PyPI.
DiffLogTest
Reproducible unittest framework for team environments — compares outputs across machines. Published on PyPI.
Diffusion Models Tutorial
Toy diffusion model on MNIST with Classifier-Free Guidance and SDS inpainting. Educational resource.
PixelCache
Seamless image data transfer between NumPy, PyTorch, and Pillow. Published on PyPI.
Photo2Panning
Convert landscape images into smooth panning videos. Simple tool for creating Ken Burns-style animations.
AdapterT2I + ControlNet
Custom Diffusers pipeline merging two conditioning methods (T2I-Adapter + ControlNet) for SDXL. Dual-conditioning for fine-grained generation control.
FCN
PyTorch implementation of Fully Convolutional Networks (VGG/ResNet) for semantic segmentation.
EmoNet
Basic emotion recognition in PyTorch. Classify facial expressions into discrete emotion categories.
Android BlinkDetection
Eye open/closed classification for Android. Real-time blink detection using mobile deep learning.