Computer-Vision-CSE527

Computer Vision Fall 2017 homeworks
HW1 - Histograms, Filters, Deconvolution, Blending
- Perform histogram equalization on color image
- Do highpass and low pass filtering in frequency domain
- Given convolution mask and a blurred image, recover original image
- Perform Laplacian pyramid blending of two images
HW2 - Panorama stitching
- Create panorama usng homographies and perspective warping on a common plane
- Compute corresponding points using SIFT featues between 2 images
- Estimate homography using corresponding points using RANSAC
- Flatten one image onto image plane of other using the computed homography
- Using cylindrical warping
- Transform image to cylindrical coordinates
- Use similar method to create cylindrical panorama
HW3 - Detection and tracking
- Implement CAMShift, Kalman Filter, Particle Filter, Optical FLow trackers
HW4 - Segmentation using SLIC and Graph cut
- Input: an image and sparse markings for foreground and background
- Calculated SLIC over image
- Calculated color histograms for all superpixels
- Calculated color histograms for FG and BG using provided sparse markings
- Constructd a graph that takes into account superpixel-to-superpixel interaction (smoothness term), as well as superpixel-FG/BG interaction (match term)
- Used graph-cut algorithm to get the final segmentation
- Created interactive UI for specifying background and foreground, and showed segmentation results
HW5 - Structured Light 3d Scanner
- Computed 2d-2d correspondence between projector and camera using binary code for each pixel
- Used 2d-2d correspondences to find 3d points
- Also added color to the estimated 3d point cloud
HW6 - Train CNN on MNIST
- Train an MNIST CNN classifier on just the digits: 1, 4, 5 and 9
- Freeze first 4 layers of this network and train on digits 0, 2, 3, 6, 8
- Use dropout and visualize learned conv layer filters