高等計算機視覺

臺北科技大學 115-1 學期課程，課號 366382。

課程大綱

This course takes a modular approach to explore core research directions in modern computer vision, covering visual recognition foundations, object detection and dense prediction, vision-language and generative models, representation learning and practical challenges, and real-world vision systems. Each module spans 2–3 weeks to allow students to genuinely engage with the problem context rather than survey topics at a surface level. The course is conducted entirely in English (EMI). Through lectures, guided paper reading, model experiments, and a final research project, students develop the ability to read papers, design experiments, and propose research directions relevant to ECE applications including edge computing, smart manufacturing, and AIoT perception.

課程進度

Week Topic 1 Course Overview and Project Guidelines 2 CNN and Transfer Learning 3 Vision Transformers 4 Object Detection I: Two-stage Methods 5 Object Detection II: One-stage, DETR, Open-vocabulary 6 Dense Prediction I: Semantic and Instance Segmentation 7 Dense Prediction II: SAM and Open-vocabulary Segmentation 8 Midterm Project Proposal 9 VLM I: CLIP and Image-Text Representation 10 VLM II: Prompting, Multimodal Reasoning, VLM Applications 11 Generative Vision: Diffusion and Synthetic Data 12 Representation Learning and Self-supervised Learning 13 Practical Issues I: Data Imbalance, Small Data, Weak Supervision 14 Practical Issues II: Robustness, Occlusion, Domain Adaptation 15 Real-world Vision Systems: Edge, Manufacturing, Robot Vision 16 Project Progress Discussion / Technical Clinic 17 Flexible Learning (Final Project) 18 Flexible Learning (Final Project)

評量方式

Assignment 1: Classification / Transfer Learning 15% Assignment 2: Detection or Segmentation 15% Assignment 3: VLM / CLIP / Generative Vision 15% Midterm Project Proposal 15% Final Project (presentation + written report) 35% Participation 5%

教材

【遵守智慧財產權觀念，請使用正版教科書，不得使用非法影印教科書】 使用外文原文書：是 • Szeliski, Computer Vision: Algorithms and Applications, 2nd ed. (free PDF available at szeliski.org) • CS231n public lecture slides and programming assignments • Hugging Face, Ultralytics, and Anomalib official documentation • Paper PDFs for all papers provided by the instructor via the course platform