Computer Vision

This repo holds all programming assignments completed for my Computer Vision course (Fall 2021).

Note: Scaffolding code was given for some of these assignments. All code beneath functions and methods is completed by me, unless otherwise noted.

Assignment Descriptions

PA1 --- NCC Image Layering

Layering of red, green and blue image filters based on a normalized cross correlation (NCC) calculation of image pyramids.

• Code: pa1_NCC_Image_Layering/pa1.ipynb

PA2 --- SIFT Feature Extraction + Matching

Implementation of SIFT feature extraction and corner matching: sobel kernels, second moment matrices, max pooling, patch descriptors, pairwise feature distances, gradient magnitudes and orientations, feature matching.

• Code: pa2_SIFT_Feature_Matching/pa2_code/pa2.ipynb
• Report: pa2_SIFT_Feature_Matching/pa2_report.pdf

PA3 --- Camera Calibration + Fundamental Matrix Estimation

Estimating camera projection and fundamental matrices using RANSAC to match images at different viewpoints.

• Code: pa3_CameraCal_RANSAC_Matrix_Estimations/pa3_code/pa3.ipynb
• Report: pa3_CameraCal_RANSAC_Matrix_Estimations/pa3_report.pdf

PA4 --- Fully Connected Neural Network (FCN)

Linear Classifier: Implementation of a linear classifier using cross entropy and SVM Hinge losses. FC Net: Implementation of a fully connected neural network. Processes implemented include forward and backward passes of affine layers, ReLU activation, softmax loss, SGD and momentum.

• Code:
  • pa4_Fully_Connected_Net/pa4_code/linear_classifier.ipynb
  • pa4_Fully_Connected_Net/pa4_code/fc_net.ipynb

PA5 --- Convolutional Neural Network (CNN)

Full naive implementation of a CNN. Naive implementations of forward and backward passes of convolution layers, batch normalization (normal and spatial) and adaptive average pooling. Network trained using cross-validation.

• Code: pa5_CNN/pa5_1_code/cnn.ipynb