Terrassa Buildings Challenge 2015

Image retrieval and classification powered by computer vision.

This project is maintained by gdsa-upc

The Terrassa Building Challenge 2015 consists in two different tasks to be solved on the Terrassa Buildings 900 dataset:


Team Retrieval (MAP) Classification (accuracy)
Egara View 0.28338 0.8000
Search Around 0.28135 0.77407
Discover Terrassa 0.20002 0.76667
Building Recognizer 0.30901 0.75556
RdE 0.26609 0.60741
Random results 0.066 0.007
What a building ! - -

(click on the team name to visit their project site containing source code and further details)

These results were assessed with two challenges in Kaggle in Class during January 2016.


The used dataset was collected by the students using only Creative Commons images. The images and annotations can be downloaded from this page from the UPC Image Processing Group.

Weekly progress

Slides Jupyter notebook
Session 1 -
Session 2 -
Session 3 Feature extraction
Session 4 Evaluation
Session 5 Retrieval
Session 6 Classification
Session 7


This project oriented course was created by Amaia Salvador and Xavier GirĂ³-i-Nieto from the ETSEIAAT, Universitat Politecnica de Catalunya, Terrassa, Catalonia