Projects & Publications

DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

ECCV, 2026 | LDR-to-HDR as generative radiance inpainting in video diffusion latent space.
Arxiv

Accelerating Video Quality Control at Netflix with Pixel Error Detection

Netflix TechBlog, 2025 | A full-resolution neural network and synthetic data pipeline to spot hot pixels.
Netflix TechBlog • Patent: US20260065628A1 / US20260065452A1 / US20260065436A1

Applied Research Projects at Apple

Apple, 2020 - 2024 | Representative research areas described at a high level due to company confidentiality:

  • Novel loss functions tailored to computer vision tasks with special requirements.
  • Efficient neural network architectures for on-device inference.
  • Low level vision models for computational photography.

SemifreddoNets: Partially Frozen Neural Networks for Efficient Computer Vision Systems

ECCV, 2020 | Fixed-topology neural networks having partially frozen weights.
Arxiv / Springer / PDF / Video • Patent: US11868892B2 / US11880763B2

A Machine Learning Imaging Core using Separable FIR-IIR Filters

ArXiv, 2020 | Fixed-function neural network hardware for efficient image processing.
Arxiv / PDF / Video • Patent: US20200074691A1

Eye Contact Correction using Deep Neural Networks

WACV, 2020 | Restores eye contact during video calls by warping and tuning the eyes.
Arxiv / PDF / Video • Press coverage: Fortune / TechXplore • Patent: US11699217B2 / US11625559B2

Seeing Through the Clouds With DeepWaterMap

Geoscience and Remote Sensing Letters, 2019
State-of-the-art convolutional neural network for pixel-wise water segmentation.
IEEEXplore / PDF / Code

VisionISP: an Image Processing Pipeline for Computer Vision Applications

ICIP, 2019 | An imaging pipeline that consists of trainable blocks.
IEEEXplore / Arxiv / PDF / Video • Patent: US11409986B2

Learning a River Network Extractor Using an Adaptive Loss Function

Geoscience and Remote Sensing Letters, 2018
A deep neural network that detects river centerlines on satellite imagery.
IEEEXplore / PDF / Code

Landmark Recognition on Moto X4 Smart Camera

Motorola Mobility, 2017 | A camera feature that detects points of interest in pictures.
Press coverage: Qualcomm Blog / Android Headlines

Surface Water Mapping by Deep Learning

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017
Pioneering deep learning model to segment water on multispectral images.
IEEEXplore / PDF / Code

RivaMap: An Automated River Analysis and Mapping Engine

Remote Sensing of Environment, 2017
Processes Landsat images, delineates rivers, and estimates their width.
ScienceDirect / PDF / Code • Media coverage: UT News

A New Database and Protocol for Image Reuse Detection

ECCV Workshop on Computer Vision for Art Analysis, 2016
Proposes a new dataset and a baseline method for detecting artworks that share common elements.
Springer / PDF


Check out my Google Scholar profile for more: Google Scholar