Chair for Computer Aided Medical Procedures & Augmented Reality
Lehrstuhl für Informatikanwendungen in der Medizin & Augmented Reality

Helisa Dhamo


Helisa Dhamo

Helisa Dhamo

  • PhD candidate

  • Email: helisa.dhamo[at]tum.de

  • Address:

Chair for Computer Aided Medical Procedures & Augmented Reality
Fakultät für Informatik
Technische Universität München
Boltzmannstr. 3
85748 Garching b. München


  • September 2020: Starting a Research Internship at Facebook Reality Labs.
  • September 2020: Gave an invited talk at ENPC on Scene Understanding and Representations.
  • February 2020: Two papers as (shared) first author accepted at CVPR '20!
  • July 2019: One paper accepted at ICCV '19!
  • May 2019: One paper accepted for publication at PRL Journal!

Student Projects

I will not be taking on new student projects from September to December 2020.


Master ThesisScene graph generation
(Helisa Dhamo Azade Farshad, Federico Tombari)
Master ThesisMeta-learning for Image Generation/Manipulation using Scene Graphs
(Azade Farshad, Helisa Dhamo, Prof. Dr. Nassir Navab)

Master ThesisRethinking Deep Learning based Monocular Depth Prediction
(Helisa Dhamo, Federico Tombari)
Bachelor ThesisComparison of methods to produce a two-layered LDI representation from a single RGB image
(Helisa Dhamo, Federico Tombari)
Master ThesisSkin Lesion Segmentation on 3D Surfaces
(Helisa Dhamo, Federico Tombari)
Master ThesisDeep Learning for Semantic Segmentation of Human Bodies
(Helisa Dhamo, Federico Tombari, Prof. Nassir Navab)



J. Wald, H. Dhamo, N. Navab, F. Tombari
Learning 3D Semantic Scene Graphs from 3D Indoor Reconstructions
IEEE Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, June 2020
The first two authors contributed equally.
H. Dhamo, A. Farshad, I. Laina, N. Navab, G. D. Hager, F. Tombari, C. Rupprecht
Semantic Image Manipulation Using Scene Graphs
IEEE Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, June 2020
The first two authors contributed equally.
H. Dhamo, N. Navab, F. Tombari
Object-driven Multi-Layer Scene Decomposition from a Single Image
International Conference on Computer Vision (ICCV), Seoul, Korea, October 2019 (bib)
H. Dhamo, K. Tateno, I. Laina, N. Navab, F. Tombari
Peeking Behind Objects: Layered Depth Prediction from a Single Image
Pattern Recognition Letters, Vol. 125, 2019 (bib)

Research Projects

My research interests include the application of deep learning in computer vision. In particular I work on ambiguous aspects of scene understanding, such as generating occluded regions and semantic manipulation.

Layered Depth Prediction

Given a single image, we regress a Layered Depth Image (LDI), which contains information about the occluded regions in the reference frame and can fill in occlusion gaps in case of small view changes.

In Peeking Behind Objects, we separate the given scene in two layers, a foreground and a background, on a pipeline based on depth prediction and object inpainting. We show the applicability of the obtained RGB-D representation in view synthesis.

In the object-driven approach, we incorporate scene understanding to decompose the scene such that it is represented as object layers as well as a layout (background) layer. This makes the number of layers that describe a scene dependent on the complexity of the scene. Additionally, besides view synthesis, it enables object removal from the original image. Project Page

Scene Graphs

Scene graphs refer to a data representation that semantically describes a scene, composed of nodes (objects) as well as edges (relationships between objects).

We address the novel problem of image manipulation from scene graphs, in which a user can edit images by merely applying changes in the nodes or edges of a semantic graph that is generated from the image. Our goal is to encode image information in a given constellation and from there on generate new constellations, such as replacing objects or even changing relationships between objects, while respecting the style from the original image. Our training strategy does not require direct supervision for image edits. Project Page | Code

We leverage inference on scene graphs as a way to carry out 3D scene understanding, mapping objects and their relationships. In particular, we propose a learned method that regresses a scene graph from the point cloud of a scene. In addition, we introduce 3DSSG, a semi-automatically generated dataset, that contains semantically rich scene graphs of 3D scenes. Project Page


May 2017 M.Sc. Computational Science and Engineering, TUM. Thesis topic: "Depth Prediction from Structured Light using Fully Convolutional Neural Networks"

Title: M.Sc.
Circumference of your head (in cm):  
Firstname: Helisa
Lastname: Dhamo
Nationality: Albania
Languages: English, German, Spanish
Groups: Computer Vision
Position: Scientific Staff
Status: Active
Emailbefore: helisa.dhamo
Emailafter: tum.de
Room: MI 03.13.041

Edit | Attach | Refresh | Diffs | More | Revision r1.45 - 17 Sep 2020 - 14:59 - HelisaDhamo

Lehrstuhl für Computer Aided Medical Procedures & Augmented Reality    rss.gif