Abdul Mannan

Academic

Career

Verbatim text

DeepScene 

DeepScene is a research project aimed towards developing a novel solution for visualizing virtual 3D 
animated scenes using text or speech. This allows the user to create 3D scenes as they please. Its 
applications include helping content creators in story-boarding, helping students visualize problems 
and ideas and can potentially be used by architects to map out building sites and visualize interior 
design concepts. 
In essence, our engine can play at the intersection of natural languages and real world objects. To 
meet the use cases in different domains, we don’t intend it to be an end to end pipeline but just a 
core which can be used for different tasks via transfer learning techniques. 
We achieve this by creating a scene graph from text or verbal input and pass that graph to a graph 
convolutional neural network for an edge classification task. This enables us to classify the graph 
edges based on distance and direction class predictions and place the new entities into the scene 
accordingly. And to bring life to the scene we further predict animations of the generated entities 
as-well. 
Our main contributions include creating scene graphs from text, creating a playground environment 
to save training examples and animating the 3D scenes. 































Technology Used:
Python, C#, Unity, ,
Supervisor Name:
Dr Omer Beg
Group Members:
Syed Zohair Abbas Hadi (I18-0671)
Abdul Mannan (I18-0577)

AI enrichment