Dr.-Ing. Hardik Jain

In this work, we propose a novel denoising technique, the icosahedral mesh denoising network (IMD-Net) for closed genus-0 meshes. A preprocessing step, exploiting the homeomorphism between genus-0 mesh and sphere, remeshes an irregular mesh using the regular mesh structure of a frequency subdivided icosahedron. Enabled by gauge equivariant convolutional layers arranged in a residual U-net, IMD-Net denoises the remeshing invariant to global mesh transformations as well as local feature constellations and orientations, doing so with a computational complexity of traditional conv2D kernel. The network is equipped with carefully crafted loss function that leverages differences between positional, normal and curvature fields of target and noisy mesh in a numerically stable fashion. In a first, two large shape datasets commonly used in related fields, ABC and ShapeNetCore, are introduced to evaluate mesh denoising. IMD-Net's competitiveness with existing state-of-the-art techniques is established using both metric evaluations and visual inspection of denoised models.

In the past few decades, the computer vision domain has achieved outstanding success in learning 3D shapes for classification, segmentation and image-based reconstruction. However, deep networks are less explored for the generative task of obtaining new 3D shapes from the learned representation. This problem becomes more prominent for 3D shapes represented as surface meshes, mainly because the mesh structure lacks regularity, an essential property for training deep generative networks. In this work, we remedy this problem by proposing a generative icosahedral mesh convolutional network (GenIcoNet) that learns data distribution of surface meshes. Our end-to-end trainable network learns semantic representations using 2D convolutional filters on the regularized icosahedral meshes. Our experiments demonstrate that both the architectures of GenIcoNet are able to outperform existing networks trained on intermediate surface mesh representations.

In this work, we try to reconstruct 3D shape from images by using a parameterized representation of the shape. We perform iterative parameterization of the surface to obtain a planar representation. This representation is encoded with surface information to generate 2D geometry images, which can be conveniently learned using traditional deep neural networks without additional overhead. Our experiments demonstrate that the proposed network learns detailed features and is able to reconstruct geometrically accurate shapes from single image.

Stereo reconstruction from image pairs is a standard method for 3D acquisition of human faces. Depending on available imagery and accuracy requirements the resulting 3D reconstructions may have deficits. In this work we remedy such deficits combining the 3D stereo reconstruction with a generic Morphable Model. Prior shape information can be obtained by already developed methods, which uses landmarks to fit a morphable model to a single image resulting in a second 3D reconstruction. This alternative to the stereo reconstruction is combined with it, allowing to prefer information from the single image reconstruction whenever the stereo reconstruction shows untypical deviations from the expected 3D features of a human face.

A method for co-registering pre-surgically acquired tomographic data with a patient at time of surgical intervention is introduced. The system using this method consists of a registration element fixed to the skin of the patient at scan and intervention time, a camera, and a laser pointing device movable on a bow rail circumferencing the body of the patient. Camera and laser bow are firmly mounted and calibrated to each other. Pre-operative planning is done using the tomographic data. After moving the patient out of the gantry, the laser ray is steered to indicate position and orientation of a linear instrument (e.g. a needle) on the patient’s skin. It is shown, that the achievable accuracy of the method presented in this paper is sufficient for periradicular therapy.

Laser bow systems are used for intervention guidance by steering laser rays. Such systems have shown applications in biopsy, nasal surgeries, drainage, pain therapy. In this paper, we introduce an image-based laser guided CT intervention. Instead of relying on traditional laser bow rigidly mounted to the CT scanner, our approach performs registration making use of a patient tracker in image as well as CT data to find the orientation of laser bow with respect to the CT data. The image is acquired by a camera rigidly connected to the laser bow. The proposed approach allows movability to the laser bow system without compromising on accuracy. Using camera allows real-time view of the intervention and can be used for futuristic applications

State of art deep neural network (DNN) models are both memory prohibitive and computationally intensive with millions of connections. Employing these models for an embedded mobile application is resource limited with large power consumption and significant bandwidth requirement. In this paper, we propose a power efficient multi-bit neural network accelerator, where we employ the technique of truncating the partial sum (PSum) results from the previous layer before feeding it into the next layer.

In this work, we treat the task for multi-view monocular imagery acquiring both body surface as well as reference markers. To fulfill the high accuracy requirements the patient is not supposed to move while images are taken. An approach towards relaxing this demanding situation is to measure small movements of the patient, e.g. with help of an electromagnetic device, and to compensate for the measured motion prior to body surface triangulation. We present two approaches for motion compensation: disparity shift compensation, and moving cameras compensation - both capable of achieving patient registration qualitatively equivalent to motion-free registration.

In this digital era, one thing that still holds the convention is a printed archive. Printed documents find their use in many critical domains such as contract papers, legal tenders and proof of identity documents. As more advanced printing, scanning and image editing techniques are becoming available, forgeries on these legal tenders pose a serious threat. Ability to easily and reliably identify source printer of a printed document can help a lot in reducing this menace.

In image forensics, detection of image forgeries involving non-linear manipulations have received a great deal of interest in recent past. Median filtering (MF) is one such non-linear manipulation technique which is quite often used in number of applications such as to hide impulse noises. Unlike other linear filtering operations, non-linear characteristics of median filtering makes it harder to detect using traditional forensics methods designed for detecting linear operations.