Internal Seminar: Prof. Jeppe Revall Frisvad, Technical University of Denmark


A big welcome to Prof. Jeppe Revall Frisvad from Technical University of Denmark
who visited our Center! Today, he presents his work on "Glass Object Appearance Acquisition Using Learning-Based Inverse Rendering". Most rendering tools have the ability to produce plausible renderings of glass objects. However, such renderings are not necessarily photorealistic. At the Technical University of Denmark, Prof. Jeppe Revall Frisvad and collaborators are studing rendering techniques for making glass object photorealistic. With that, they can generate data set of synthetic images of glass objects with corresponding reference data. A large set of photorealistic synthetic images is useful for deep learning of the relation between glass appearance in a photograph and the shape and optical properties of the object. Prof. Jeppe Revall Frisvad gives us some indications of the performance of a convolutional neural network in acquiring the shape of a glass object from a photograph of it “in the wild”. We are greatful to prof. Jeppe Revall Frisvad, as his talk has activated discussion on how such an approach could be used for more accurate scanning of glass object shape when photographed in a robot-controlled light-camera configuration.

Internal Seminar: M. E. Brix Doest, Technical University of Denmark

Mads Emil Brix Doest from DTU presents today his research on Single-Shot Analysis of Refractive Shape Using Convolutional Neural NetworksThe appearance of a transparent object is determined by a combination of refraction and reflection, as governed by a complex function of its shape as well as the surrounding environment. Prior works on 3D reconstruction have largely ignored transparent objects due to this challenge, yet they occur frequently in real-world scenes. Brix Doest presents an approach to estimate depths and normals for transparent objects using a single image acquired under a distant but otherwise arbitrary environment map. In particular, they use a deep convolutional neural network (CNN) for this task. Unlike opaque objects, it is challenging to acquire ground truth training data for refractive objects, thus, they propose to use a large-scale synthetic dataset. To accurately capture the image formation process, they use a physically-based renderer and demonstrate that a CNN trained on our dataset learns to reconstruct shape and estimate segmentation boundaries for transparent objects using a single image, while also achieving generalization to real images at test time. In the presented experiments, they extensively study the properties of our dataset and compare to baselines demonstrating its utility.

  • 1
  • 2