DLCV 2018 Team 1
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.DS_Store
.gitignore
Readme.md
STN_streetview.py
STN_traffic_signals.py
STResults1.png
STResults2.png
STTable1.png
STmodel.png
SVHN_STN.py
spatial_transformer.ipynb
spatial_transformer.py
street_view.py

Readme.md

Team 1 - DLCV Summer School - UPC

This repository is created as a final project for the Deep Learning for Computer Vision summer school at UPC, Barcelona.

Team 1 is formed by Armand Comas, Victor Aubin & Guillem París

Spatial Transformer Net

Original code from Pytorch tutorial: https://pytorch.org/tutorials/intermediate/spatial_transformer_tutorial.html

Provided code's goal is to learn how to perform spatial transformations of the input image in order to preserve, as much as possible, the geometrical invariance of the model.

ST Model

Results with Datasets

ST Model

ST Model

Applying STNet to distorted images improved the classification results by 3% with MNIST dataset & 5% with GTSRB dataset.

ST Model