AwesomeLearning

Good courses, tutorials and websites to learn from.

Courses

• CMSC 733: Classical and Deep Learning Approaches for Geometric Computer Vision at UMD: The only course in the world which requires students to solve each project with both classical and deep learning approach.

  • Homework 0: Pb lite boundary detection
  • Homework 1: Camera calibration
  • Project 1: Panorama stiching + Supervised and Unsupervised Deep Homography
  • Project 2: Face Swap + Position Map Regression Network
  • Project 3: Structure from Motion
  • Project 4: SfM Learner

• CMSC 426: Computer Vision at UMD: Arguably the hardest computer vision undergraduate course in the world.

  • Homework 1: Linear Least Squares
  • Project 1: Color Segmentation using GMM
  • Project 2: Panorama Stitching
  • Project 3: RotoBrush (tracking felixble boundaries)
  • Project 4: Factor graph (GTSAM) based SLAM

• CMSC 828T: Vision, Planning and Control in Aerial Robotics at UMD: UMD's version of MEAM 620 course at UPenn. Added Factor graph (GTSAM) based SLAM.

  • Project 1: Implementation of A* based path planner, trajectory parametrization and following in simulation
  • Project 2: Factor graph (GTSAM) based SLAM
  • Project 3: Flying through a known window using Parrot ARDrone 2

• ENAE 788M: Hands On Autonomous Aerial Robotics: Hands on and competition targetted version of CMSC 828T.

  • Project 1a: Attitude estimation using Madgwick Filter
  • Project 1b: Attitude estimation using Unscented Kalman Filter
  • Project 2: Trajectory Following on the PRG Husky (Modified Parrot Bebop 2)
  • Project 3a: Flying through known colored windows
  • Project 3b: Detecting and landing on a circular bullseye target
  • Project 4a: Stereo Visual Odometry
  • Project 4b: Obstacle avoidance from front-facing monocular camera using pose form down-facing stereo visual odometry
  • Project 5: The final race

• MEAM 620: Advanced Robotics at UPenn: The full fledged version of the Coursera's Robotics: Aerial Robotics course. Covers concepts of quadrotor dynamics, controls and computer vision needed for quadrotor autonomy. The projects are given below:

  • Project 1: Implementation of A* based path planner, trajectory parametrization and following in simulation and on hardware.
  • Project 2: April Tag based pose estimation (given tag positions in the world frame), then using EKF to estimate velocities using optical flow AND SLAM with bundle adjustment (2016 offering) along with replacement of EKF by ESKF.
  • Project 3 (2015 offering): Implementation of CAPT and D-CAPT for multi-robot assignment OR implementing project 2 on hardware OR estimation and control of Parrot Bebop.

• CIS 581: Computer Vision and Computational Photography at UPenn: The full fledged version of the MIT edx's Robotics: Vision Intelligence and Machine Learning course. Covers concepts of convolution, edge detection, triangulation, TPS, RANSAC, seam carving, feature extraction and detection

  • Project 1: Canny Edge detection
  • Project 2: Image Morphing
  • Project 3: Image Mosaicing/Panorama Stitching
  • Project 4: Face Replacement

• CIS 580: Machine Perception at UPenn: The full fledged version of the Coursera's Robotics: Perception course. Covers concepts of 3D multi-view geometry including camera calibration, epipolar geometry, vanishing points, PnP, triangulation and Bundle Adjustment. Different variations of the following projects can be found on the 2019 offering of the course.

  • Homework 1: Dolly Zoom, Vanishing Points and Homography
  • Homework 2: Single View Meterology
  • Project 3: Camera Calibration
  • Project 4: Structure from Motion

• ESE 650: Learning In Robotics at UPenn: The full fledged version of the Coursera's Robotics: Estimation and Learning course. Covers concepts of baeysian filtering for various applications.

  • Project 1: Gaussian Mixture Models based color segmentation for Barrel Detection
  • Project 2: Unscented Kalman Filter based orientation estimation for panorama stitching
  • Project 3: Hidden Markov Models for Guesture recognition using IMU data
  • Project 4: SLAM on LIDAR Data using Particle Filters and Scan Matching
  • Project 5: Imitation Learning to learn costs for path planning
  • Project 6: ICP on RGB-D data for object reconstruction OR Plane Fitting on RGB-D data using Mean-shift OR Fast IK planning of robotic arms

• AUTONAVx: Autonomous Navigation for Flying Robots: A toned down but more hands-on version of the MEAM 620 course at UPenn.

  • Refer for good and simple to understand slides for vision on quadrotors

• CSCI 5980: Multiview 3D Geometry in Computer Vision at UMN: Different variant of CIS 580 course at UPenn.

  • Refer for more polished slides (math explained more clearly with better figures) than CIS 580

Tutorials

Blogs