PurdueX: Computer Vision for Embedded Systems

i. Use computer vision to analyze images.

ii. List the constraints of embedded systems.

iii. Explore design space of computer vision.

iv. Evaluate different methods for accuracy/time tradeoffs.

Lecture topics:

• Overview, image data formats, OpenCV
• Edge detection and segmentation
• Applications of computer vision in embedded systems
• Datasets, bias, privacy, competitions
• Machine learning and PyTorch
• Performance and resources (time, memory, accuracy)
• Object detection and motion tracking
• Data annotation and generation
• Quantization
• Pruning and network architecture search
• Tree modular networks
• Vision in context, MobileNet
• Real-time scheduling

Fall 2021 course feedback:

• The organization of the content is superb.
• It was a very innovative class. It was refreshing that this class was focused on learning, rather than only testing the students.
• I think the concepts are delivered very well.
• I enjoyed having the exposure to quantization.
• The project structure was great in my opinion.

Spring 2022 course feedback:

(Dr. Lu gave the short course at the Seoul National University in South Korea)

• Instruction for the assignments is clear and explicit, and these assignments help me to fully understand the contents learned in the lecture.
• Lectures were very well done.
• I can feel the professor prepared the lecture well.
• Good quiz, good lecture, good lecturer
• Everything was perfect.

Q: Does this course focus on theory or practice?

A: The emphasis is using machine learning, not about deriving equations for the theory of machine learning. For example, we will use the tools in PyTorch.