Project Work at DLCV
|Dídac Surís||Francesc Busquet||Ricardo García||Francisco Herranz|
The first task consists on the creation of a simple neural network to solve the classification task of the dataset CIFAR10, where we explored the variance and bias trade-off by changing some parameters and hyperparameters of our model.
The second task consists on building a soft-max classifier on top of a pre-trained model such as CIFAR10 or off the shelf imagenet model. This is done for the Terrassa data Set.
The third task consists on applying some ideas such as transfer learning/fine tunning to build a powerful classifier for the Terrassa Data Set.
The fourth tasks uses the previous ideas and techniques to improve the classification model by using an intermediate step where we trained/fine tunned the net for the oxford buildings task, so our classifier for the Terrassa data set can learn the features of buildings in a more precise way.
In the fifth task we generated an Image Classifier using the Inception V3 model, furthermore we generated a user interface to allow the users upload images while the system classifies them showing the class label and its predicted probability. To be able to execute this task it is needed to have R and the Jupyter R kernel installed, as well as the packages used in it.
You can access our web through: https://telecombcn-dl.github.io/2017-dlcv-team1/