CSE 252D: Advanced Computer Vision (original) (raw)
CSE 252D: Advanced Computer Vision, Spring 2021
Instructor: Manmohan Chandraker
Email: mkchandraker [AT] eng [DOT] ucsd [DOT] edu
Lectures: WF 5-6:20pm on Zoom
Instructor office hours: Thu 3-4pm on Zoom
TA: Yu-Ying Yeh (yuyeh@eng.ucsd.edu)
TA office hours: Mon 10-11am on Zoom
Class discussion and message board: Piazza
Overview
This course will cover advanced concepts in computer vision. Example topics include 3D reconstruction, face recognition, object detection, semantic segmentation and domain adaptation.
Prerequisites
This is an advanced class, covering recent developments in computer vision and will extensively refer to papers. Prior background in computer vision and machine learning is required, through research experience or as covered by CSE 252A, 252B, 250B and similar offerings. Students are encouraged to contact the instructor if unsure about meeting any criteria for enrollment.
Course Format and Requirements
This will be a lecture-based course in which the majority of the material will be covered by the instructor. Students will also give a short presentation on an assigned paper (there will be an asynchronous option). Besides, the class will have three assignments and a final exam.
Grades will be weighted as 60% for assignments, 10% for presentation and 30% for the final exam. The goal of the course is to understand the current state of computer vision and gain appreciation of its limits and potential.
Topics
The course will cover a diverse range of topics in computer vision, including:
- Feature detection and matching
- Optical flow
- Structure from motion
- Face recognition
- Human pose estimation
- Material and lighting
- Semantic segmentation
- Object detection
- Action recognition
- Domain adaptation
Outline
Mar 31: Introduction
- Lecture [PDF] Apr 02: Overview
- Lecture [PDF]
- References:
- Solutions [PDF] Apr 07: Metric Learning
- Lecture [PDF]
- References:
- Large Margin Nearest Neighbors
- Do Convnets Learn Correspondence? Apr 09: Learning correspondence
- Lecture [PDF]
- References:
- Lecture [PDF]
- Pyramidal Implementation of the Affine Lucas-Kanade Feature Tracker Apr 16: Learning optical Flow
- Lecture [PDF]
- References:
- FlowNet: Learning Optical Flow with Convolutional Networks
- Optical Flow Estimation Using a Spatial Pyramid Network Apr 21: Structure from Motion
- Lecture [PDF]
- References:
- Fundamental matrix
- RANSAC Apr 23: Learning Structure from Motion
- Lecture [PDF]
- References:
- SFM-Net Apr 28: Practical Structure from Motion
- Lecture [PDF]
- References:
- Five Point Relative Pose
- Three Point Absolute Pose (can ignore algebraic details)
- ORB-SLAM
- Bundle Adjustment Quiz 2
- Solutions [PDF] Apr 30: Face Recognition: I
- Lecture [PDF]
- References:
- DeepFace
- DeepID2
- LFW
- MS-Celeb-1M May 05: Face Recognition: II
- Lecture [PDF]
- References:
- FaceNet
- Center Loss
- SphereFace, CosFace, ArcFace May 07: Human Pose Estimation: I
- Lecture [PDF]
- References:
- DeepPose
- Convolutional Pose Machines
- Stacked Hourglass Networks May 12: Human Pose Estimation: II
- Lecture [PDF] May 14: Semantic segmentation: I
- Lecture [PDF]
- References:
- Solutions [PDF] May 19: Semantic segmentation: II
- Lecture [PDF]
- References:
- Dilated convolutions
- Dilated Residual Networks
- PSPNet May 21: Object detection: I
- Lecture [PDF]
- References:
- R-CNN
- Fast R-CNN May 26: Object detection: II
- Lecture [PDF]
- References:
- Faster R-CNN May 28: Object detection: III
- Lecture [PDF]
- References:
- Mask R-CNN
- SSD Quiz 4
- Solutions [PDF] Jun 02: Domain adaptation
- Lecture [PDF]
- References:
- ADDA Jun 04: Review
- Lecture [PDF] Jun 05: Supplement on t-SNE
- Lecture [PDF]
- References:
Resources
- Books: There are no books required for this course. Any chapters of books that are extensively referenced in class will be provided as hand-outs.
- Papers: The papers will be provided as PDFs or made available for download through provided links.
Last modified: Sun, Mar 28, 2021