Human Perceptual-Centric Vision and Multimodal Representation Learning (2026.08.28 Updated)
Human Perceptual-Centric Vision and Multimodal Representation Learning (2026.08.28 Updated)
[Syllabus] (on E3)
Textbook Download Link:
[Computer Vision: Algorithms and Applications, 2nd ed.]
[Martin J. Tovée, An Introduction to the Visual Systems (2nd ed.), Cambridge University Press, 2012] (login via NYCU Library portal)
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 1 Introduction to Perception
重點讀:definition of perception, perceptual process, physiological approach, psychophysical approach, cognitive influences on perception.
目的:建立 perception 的基本概念與研究方法。
2. Szeliski, Computer Vision: Algorithms and Applications, Ch. 1 Introduction
目的:建立 computer vision 如何定義 machine perception。
3. (Supplementary) Tovée, An introduction to the visual system, Ch. 1 Introduction
4. (Supplementary) Kingdom & Prins, Psychophysics: A Practical Introduction, Ch. 1.1 What is Psychophysics?
目的:預告後續如何量測人類感知、threshold、JND 與 preference。
Pre-class questions:
1. What is the difference between sensing and perceiving?
2. Does a neural network “see” an image, or does it only transform pixels?
3. What makes a representation good?
4. Can a model be semantically correct but perceptually wrong?
5. In what applications is human perceptual alignment more important than classification accuracy?
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 3 The Eye and Retina, Ch. 4 The Visual Cortex and Beyond
2. (Supplementary) Tovée, An introduction to the visual system, Ch. 2 The eye and forming the image, Ch. 4 The organisation of the visual system
Pre-class questions:
1. In what ways is the human eye similar to a camera, and in what ways is it fundamentally different?
2. Why does the fovea matter for visual quality and VR rendering?
3. Why does the visual system emphasize contrast rather than absolute brightness?
4. What is a receptive field, and why is it important for representation learning?
5. How is the visual hierarchy similar to, and different from, a CNN?
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 2 Introduction to the Physiology of Perception
2. Tovée, An introduction to the visual system, Ch. 2 The eye and forming the image
3. (Supplementary) Szeliski, Computer Vision: Algorithms and Applications, Ch. 2 Image Formation
4. (Optional) 選讀論文:CSF / temporal CSF / display perception / perceptual coding
Pre-class questions:
How can we think of human vision as a signal or information processing system?
我們可以如何把人類視覺理解成一個訊號/資訊處理系統?
What is visual contrast, and why might contrast matter more than raw brightness?
什麼是視覺對比?為什麼對比可能比絕對亮度更重要?
What is spatial frequency? What is the difference between low and high spatial frequency information?
什麼是空間頻率?低空間頻率與高空間頻率資訊有什麼差異?
Do humans have the same sensitivity to all spatial and temporal frequencies? Why or why not?
人眼對所有空間與時間頻率都一樣敏感嗎?為什麼?
How might contrast sensitivity influence display design, video compression, or perceived video quality?
對比敏感度可能如何影響顯示設計、視訊壓縮或主觀視訊品質?
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 9 Perceiving Color
2. (Supplementary) Szeliski, Computer Vision: Algorithms and Applications, Ch. 2 Image Formation
3. (Supplementary) Tovée, An introduction to the visual system, Ch. 3 Retinal colour vision, Ch. 7 Colour constancy
4. (Optional) 選讀論文:YUV / YCbCr / chroma subsampling / chroma CSF
Pre-class questions:
1. Does a wavelength inherently “have” a color? Why or why not?
If 650 nm light is usually perceived as red, what role does the visual system play in creating that experience?
2. Why can humans perceive many different colors with only three types of cones?
Think about how the relative responses of S-, M-, and L-cones may encode color.
3. Why are both trichromatic theory and opponent-process theory needed to explain human color vision?
What does each theory explain at a different stage of visual processing?
4. Why can the same physical color appear different under different backgrounds or illumination conditions?
Consider simultaneous color contrast, chromatic adaptation, and color constancy.
5. Why do image and video systems often preserve luminance detail more carefully than chrominance detail?
What property of human vision makes YCbCr and chroma subsampling (e.g., 4:2:0) perceptually effective?
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 4 The Visual Cortex and Beyond
2. (Supplementary) Szeliski, Computer Vision: Algorithms and Applications, Ch. 4 Image Processing, Ch. 5 Deep Learning
3. (Supplementary) Tovée, An introduction to the visual system, Ch. 5 Primary visual cortex
4. (Optional) 選讀論文:Gabor filters / energy models / CNN visualization
Pre-class questions:
What does it mean for a neuron to be selective or tuned?
Does it respond to only one stimulus, or to a range of feature values?
Which properties of an edge could different neurons represent?
Consider orientation, position, spatial frequency, contrast, and motion.
Why are sinusoidal gratings useful for studying visual neurons?
What can they isolate more easily than natural images?
Why might early CNN filters resemble V1 receptive fields?
Does visual similarity prove that the underlying mechanisms are identical?
Can equally sharp video frames produce different perceived quality?
Consider frame rate, irregular timing, rebuffering, and interaction latency.
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 1 Introduction to Perception
2. Tovée, An introduction to the visual system, Ch. 6 Visual development: an activity-dependent process
3. (Supplementary) Kingdom & Prins, Psychophysics: A Practical Introduction, Ch. 1–3(方法觀與知覺測量框架)
4. (Optional) 選讀論文:ideal observer / Bayesian perception / neural adaptation / efficient coding
Pre-class questions:
If perception improves after practice, what changed?
Does learning reshape sensory features—or only decision weights?
How should prior experience influence noisy sensory evidence?
Can self-supervised learning model human perceptual development?
Readings:
1. Kingdom & Prins, Psychophysics: A Practical Introduction, Ch. 2–8 為主
- dichotomous classification schemes
- performance- vs appearance-based procedures
- psychometric functions
- adaptive methods
- signal detection measures
2. Goldstein & Cacciamani, Sensation and Perception, Appendix A Methods of Adjustment and Constant Stimuli
3. (Optional) 選讀論文:ROC / SDT / threshold estimation
Pre-class questions:
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 2 Basic Principles of Sensory Physiology, Ch. 5 Perceiving Objects and Scenes
2. Kingdom & Prins, Psychophysics: A Practical Introduction, 回看 threshold / masking / psychometric measurement 相關章節
3. (Optional) 選讀論文:SSIM, LPIPS, DISTS, VMAF, IQA/VQA survey
Pre-class questions:
Readings:
1. Szeliski, Computer Vision: Algorithms and Applications, Ch. 5 Deep Learning, Recognition
2. (Optional) 選讀論文:CLIP, BLIP, BLIP-2, LLaVA, ALIGN, Flamingo
Pre-class questions:
Readings:
1. (Supplementary) Goldstein & Cacciamani, Sensation and Perception, perception / attention / object understanding
2. (Supplementary) Kingdom & Prins, Psychophysics: A Practical Introduction, psychophysical measurement 與 scaling
3. (Optional) preference learning, RLHF/RLAIF, personalization, fairness in multimodal systems, human evaluation protocols
Pre-class questions:
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 6 Visual Attention
2. (Supplementary) Szeliski, Computer Vision: Algorithms and Applications, Ch. 5 Deep Learning
3. (Supplementary) Tovée, An introduction to the visual system, Ch. 8 Object perception and recognition(含 visual attention and working memory)
4. (Optional) 選讀論文:saliency, eye tracking, Grad-CAM, attention rollout, relevance maps
Pre-class questions:
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 8 Perceiving Motion
2. (Supplementary) Tovée, An introduction to the visual system, Ch. 10 Motion perception
3. (Optional) 選讀論文:temporal CSF, flicker fusion, frame rate perception, motion blur, VR/AR temporal quality
Pre-class questions:
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 1 Introduction to Perception, Ch. 4 The Visual Cortex and Beyond
2. Tovée, An introduction to the visual system, Ch. 12 What is perception?
3. (Supplementary) Szeliski, Computer Vision: Algorithms and Applications, Ch. 5 Deep Learning
4. (Optional) 選讀論文:representation similarity, brain-score, adversarial examples, human-vs-model comparison
Pre-class questions:
Readings:
1. Goldstein & Cacciamani, Sensation and Perception, Ch. 8 Perceiving Motion, Ch. 10 Perceiving Depth and Size
2. Tovée, An introduction to the visual system, Ch. 9 Face recognition and interpretation, Ch. 11 Brain and space
3. (Optional) 選讀論文:stereo, motion in VR, avatar perception, uncanny valley, human likeness, telepresence quality
Pre-class questions:
每組至少引用:
- 1 本 perception / vision science 教材章節
- 2–4 篇近五年論文
- 1 個實驗或 evaluation protocol 來源