Muhammad Naufil

Developing data processing pipelines to transform raw camera and LiDAR ROS bag recordings into structured, ML-ready datasets for autonomous driving model training.

Understanding sign language and co-speech gestures through phonemes

• Finetuning vision foundation models on a huge collection of deepfakes with access to clusters of 8xA100 GPUs. • Large-scale generation of deepfakes using state-of-the-art generative open and closed source models • Learned that features distribution for every generative model is widly different and thus requires us finetune our deepfake detection model for every new generative model that comes out to learn its unique distribution

Developed a one-shot computer vision system for retail shelf monitoring; evaluated SuperGlue and LightGlue to recognize products from a single reference image and detect out-of-stock items without retraining for frequent packaging changes.

3D Human Pose Estimation: • Created MoCap product with a single colleague for a $150k yearly contract with Siemens. • Extended open-source monocular synthetic dataset of 3D motion generation pipeline to multi-view using Blender • Created a mini motion capture (MoCap) system using two RGB-D cameras Human Action Recognition / Segmentation: • Created the demo for the SBIR grant from NASA’s Human Research Program (HRP). • Ran pilot projects with Honda, Suzuki, Nissan