Skip to content

Implementation of visual search using Bags of Visual Words with SIFT detectors. Implementation of Neural Networks for object recognition. Labs and final projects from the corresponding ENS Ulm courses by Jean Ponce, Ivan Laptev, and Cordelia Schmidt

Notifications You must be signed in to change notification settings

rmissaoui/Computer_Vision_And_Object_Recognition

Repository files navigation

Computer Vision and Object Recognition

In this repository, you'll find the labs and final projects from the corresponding ENS Ulm course by Pr. Jean Ponce, Ivan Laptev, and Cordelia Schmidt, 2019 edition.

The topics covered are the following ones:

  1. Lab1: Introduction to Local Invariant Features using Scale Invariant Feature Transform: SIFT points of interests, image matching using Bags of Visual Words: BOVW. Given a picture of a portion of a painting and a database of scans of paintings, identify the original picture.

  2. Lab2: Introduction to Convolutional Neural Network: convolution operation, backpropagation of the gradient, pooling, multiple layers, drops, non-linear activation functions, training on CIFAR 10 and MNIST.

  3. Final project: Studying and experimenting with CycleGANs. It's about style transfer and high level properties transfer from a set of images to another one. Example: convert any image of a landscape taken in the summer to its winter counterpart. In this study, I investigate a pytorch implementation of the cycleGAN: I reproduced some of the results of the paper, applied the CycleGAN on two new datasets for real face to cartoon and sad to happy image translation. I also played with the loss functions to highlight the importance of each loss member and its effect.

About

Implementation of visual search using Bags of Visual Words with SIFT detectors. Implementation of Neural Networks for object recognition. Labs and final projects from the corresponding ENS Ulm courses by Jean Ponce, Ivan Laptev, and Cordelia Schmidt

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published