ArchVis

What it does

ArchVis (short for "Architecture Vision") tells you a building's exact architectural style (e.g. postmodern, neoclassical, Victorian, etc.) via your phone's camera.

Inspiration

A strong pre-existing interest in architecture and computer vision led us to our initial idea. But it was our overarching desire to decompose and use object recognition that motivated us to persist through the technical and psychological challenges of this project.

How we built it

(1) trained the IBM classifier to recognize prevalent architectural styles existing in North America (2) built the framework for the classifier to use its new knowledge to interpret inputted images of architecture (3) implemented mobile app, allowing users to input images they take with their phone-camera

Challenges we ran into

Technical barriers like (1) quality loss in input-photos due to compression, (2) positive-ratings for non-building objects; but also a shortage of programming experience that required the application creation-process and learning-process to happen concurrently.

Logo made with logomakr.com

Accomplishments that we're proud of

As two architecture enthusiasts, we successfully used a cutting-edge technology for a purpose meaningful to our interests; as well as those of the greater architecture community. With the completion of this project, we feel more confident in our ability to develop apps and complex APIs.

What we learned

(1) Bluemix Basics - deployed a sample Java program to the cloud (2) Creating new environment variables - got Maven on computer (3) Capturing and saving a photo with Android Studio (4) Multi-threading Android apps in order to stuff on the web. (5) How to use a general-purpose machine learning API to build something we were passionate about.

What's next for ArchVis

In the spirit of the API it's based on (Watson), we want users to be able to contribute to ArchVis's growth in accuracy by implementing a self-learning algorithm in ArchVis that uses photos (tagged with their respective style(s) by users) as training material. We would also like to add the following features:

(1) a dynamic explanation of why each inputted-building corresponds to its outputted-style (i.e. shape, roof, columns) (2) the ability for users to "favorite" and keep track of the landmarks he/she has visited

--

Logo made with logomakr.com

Built With

• acer
• android-studio
• dell
• ibm-watson
• java
• logomakr
• multithread