Software

• I've created a web app that shows users what they look like without hair.
• Powered by a compact diffusion model that generates realistic bald transformations from uploaded images.
• The model is compressed and quantized for efficient deployment on a small ARM server. Free to use for everyone :-)

• A simple (easy to implement, fast) online tracking algorithm for multiple objects in videos.
• The algorithm combines Kalman filter motion modeling and a deep learning model for appearance feature extraction.
• This is the official code for the paper Simple Online and Realtime Tracking with a Deep Association Metric.

• Code to train a feature space for person re-identification and object tracking, using a reparametrization of the softmax classifier.
• This code has been used to train the appearance descriptor of Deep SORT.
• It is the official code for the paper Deep Cosine Metric Learning for Person Re-Identification.

Network flow formulations provide a principled framework for solving the data association problem in multiple object tracking, enabling globally optimal solutions under well-defined assumptions.

This library offers several solvers for the min-cost flow problem, including an efficient C++ implementation of the k-shortest paths algorithm, which leverages structural properties specific to multi-object tracking to improve computational efficiency.