HIT × Samsung Student Projects
Three joint projects in camera imaging and artificial intelligence, run by the Holon Institute of Technology together with Samsung. Each project is a full-year final project for a team of up to three students.
- Projects
- 3
- Credit
- 7 points each
- Team size
- Up to 3 students per project
- Duration
- Two semesters
About the program
Research projects. All three projects are research projects for Samsung. There is no ready recipe: the team defines the details of the method, reads the literature, tries ideas, measures, and learns from what does not work. Each project requires independent work, proactive exploration and a lot of trial and error, with regular meetings with the supervisor and periodic meetings with the Samsung team.
Students work on real problems from the camera industry, with guidance from an academic supervisor at HIT and periodic meetings with the Samsung team. All three projects deal with how cameras capture images and how image quality is measured, so teams can share data, tools and ideas.
- What you get: experience with modern deep learning, camera technology and research methods, a working system, and a strong portfolio project.
- How it runs: regular meetings with the supervisor, periodic meetings with the Samsung team, and a final presentation and report.
- Who it suits: students with good Python skills. Projects 1 and 2 require prior knowledge of deep learning and PyTorch; Project 3 requires prior knowledge of computer vision or image processing. This knowledge should come from a completed course or a significant project. Each project page lists the full prerequisites.
The projects
Patch Quality Scanner
A deep learning model that scans an image in small patches and scores each one for overall quality and for five distortion families: geometry, color, light, contrast and focus. Trained on images with synthetic distortions of known type and strength.
Deep learning · Image qualityRAW Sensor Digital Twin
A model that predicts the RAW Bayer mosaic a camera sensor would record, before demosaicing, from an image, the capture conditions (exposure, ISO, aperture, light) and the sensor and lens parameters. Combines camera physics with deep learning.
Deep learning · Camera physicsSelf-Driving Camera Lab
Software that remotely controls a camera test lab (lights, motors, stages, the camera under test), sets itself up automatically, and searches test conditions for cases where the camera under-performs.
Computer vision · Robotics