Publications de Radhakrishna Achanta

Exploiting the Signal-Leak Bias in Diffusion Models

Sabine Süsstrunk, Radhakrishna Achanta, Mahmut Sami Arpa, Athanasios Fitsios, Martin Nicolas Everaert

There is a bias in the inference pipeline of most diffusion models. This bias arises from a signal leak whose distribution deviates from the noise distribution, creating a discrepancy between training and inference processes. We demonstrate that this signa ...

2024

Partial annotations in active learning for semantic segmentation

Radhakrishna Achanta, Katrin Beyer

Semantic segmentation with deep learning plays a crucial role in various fields, including civil engineering, particularly in tasks such as damage assessment and urban planning. This paper addresses the challenge of efficiently training deep learning model ...

2024

Digital Twins of Stone Masonry Buildings for Damage Assessment

Radhakrishna Achanta, Katrin Beyer, Bryan German Pantoja Rosero

Digital twins are virtual models of physical objects or systems that enable real-time monitoring and analysis. In the ﬁeld of stone masonry buildings, digital twins can be used to assess damage, predict maintenance needs, and opti- mize building performanc ...

Springer2024

Diffusion in Style

Sabine Süsstrunk, Radhakrishna Achanta, Mahmut Sami Arpa, Martin Nicolas Everaert

We present Diffusion in Style, a simple method to adapt Stable Diffusion to any desired style, using only a small set of target images. It is based on the key observation that the style of the images generated by Stable Diffusion is tied to the initial lat ...

2023

Damage-augmented digital twins towards the automated inspection of buildings

Radhakrishna Achanta, Katrin Beyer, Bryan German Pantoja Rosero

Current procedures for the rapid inspection of buildings and infrastructure are subjective, time-consuming, and cumbersome to document, necessitating new technologies to automate the process and eliminate these shortcomings. Fortunately, recent development ...

2023

Towards real-time digital twins for post-earthquake damage assessment of masonry buildings

Radhakrishna Achanta, Katrin Beyer, Bryan German Pantoja Rosero

Currently, post-earthquake assessment is performed visually, which implies that inspectors visit the damaged buildings and, based on their expertise, assign a damage grade. Yet, visual inspection is subjective, time-consuming and difficult to document. In ...

2023

Automated image-based generation of finite element models for masonry buildings

Radhakrishna Achanta, Katrin Beyer, Bryan German Pantoja Rosero

To predict the response of masonry buildings to various types of loads, engineers use finite element models, specifically solid-element and macro-element models. For predicting masonry responses to seismic events in particular, equivalent frame models-a su ...

SPRINGER2023

VETIM: Expanding the Vocabulary of Text-to-Image Models only with Text

Sabine Süsstrunk, Radhakrishna Achanta, Mahmut Sami Arpa, Martin Nicolas Everaert

Text-to-image models, such as Stable Diffusion, can generate high-quality images from simple textual prompts. With methods such as Textual Inversion, it is possible to expand the vocabulary of these models with additional concepts, by learning the vocabula ...

BMVA2023

Digital twins for post-earthquake damage assessment of free-standing buildings including damage information

Radhakrishna Achanta, Katrin Beyer, Amir Rezaie, Bryan German Pantoja Rosero

We present a pipeline for the automatic generation of digital twins of free-standing buildings for damage assessment. Our method takes multiple view images of the target building as input and processes them to produce a 3D model in levels of detail (LOD) f ...

2022

Generating LOD3 building models from structure-from-motion and semantic segmentation

Pascal Fua, Radhakrishna Achanta, Katrin Beyer, Mateusz Kozinski, Fernando Perez Cruz, Bryan German Pantoja Rosero

This paper describes a pipeline for automatically generating level of detail (LOD) models (digital twins), specifically LOD2 and LOD3, from free-standing buildings. Our approach combines structure from motion (SfM) with deep-learning-based segmentation tec ...

2022

TOPO-Loss for continuity-preserving crack detection using deep learning

Pascal Fua, Radhakrishna Achanta, Katrin Beyer, Fernando Perez Cruz, Doruk Oner, Bryan German Pantoja Rosero

We present a method for segmenting cracks in images of masonry buildings damaged by earthquakes. Existing methods of crack detection fail to preserve the continuity of cracks, and their performance deteriorates with imprecise training labels. We address th ...

2022

Machine-learning for damage assessment of rubble stone masonry piers based on crack patterns

Radhakrishna Achanta, Amir Rezaie

Under seismic actions, stone masonry buildings are prone to damage. To assess the severity of damaged masonry buildings and their failure modes, engineers connect these problems to surface crack features, such as the crack width and the extent of cracking. ...

ELSEVIER2022

Adaptive extreme video completion

Sabine Süsstrunk, Radhakrishna Achanta, Ruofan Zhou, Majed El Helou

A computer-implemented video completion method is proposed for reconstructing a sparsely sampled video comprising sparsely sampled video frames comprising sparse sets of picture element intensity values. The method comprises: distributing the sparse sets o ...

2021

Height And Weight Estimation From Unconstrained Images

Radhakrishna Achanta, Ntorina Thanou

We address the difficult problem of estimating the attributes of weight and height of individuals from pictures taken in completely unconstrained settings. We present a deep learning scheme that relies on simultaneous prediction of human silhouettes and sk ...

IEEE2020

Comparison of crack segmentation using digital image correlation measurements and deep learning

Radhakrishna Achanta, Katrin Beyer, Amir Rezaie, Michele Godio

Reliable methods for detecting pixels that represent cracks from laboratory images taken for digital image correlation (DIC) are required for two main reasons. Firstly, the segmented crack maps are used as an input for some DIC methods that are based on di ...

2020

Scale-Adaptive Superpixels

Pascal Fua, Sabine Süsstrunk, Radhakrishna Achanta, Pablo Marquez Neila

Size uniformity is one of the prominent features of superpixels. However, size uniformity rarely conforms to the varying content of an image. The chosen size of the superpixels therefore represents a compromise - how to obtain the fewest superpixels withou ...

Society for Imaging Science and Technology2018

Deep Feature Factorization for Concept Discovery

Sabine Süsstrunk, Radhakrishna Achanta, Edo Collins

We propose Deep Feature Factorization (DFF), a method capable of localizing similar semantic concepts within an image or a set of images. We use DFF to gain insight into a deep convolutional neural network's learned features, where we detect hierarchical c ...

SpringerLink2018

Deep Residual Network for Joint Demosaicing and Super- Resolution

Sabine Süsstrunk, Radhakrishna Achanta, Ruofan Zhou

The two classic image restoration tasks, demosaicing and super-resolution, have traditionally always been studied indepen- dently. That is sub-optimal as sequential processing, demosaic- ing and then super-resolution, may lead to amplification of ar- tifac ...

Society for Imaging Science and Technology ( IS&T )2018

Fourier-Domain Optimization for Image Processing

Sabine Süsstrunk, Radhakrishna Achanta, Frederike Dümbgen, Majed El Helou

Image optimization problems encompass many applications such as spectral fusion, deblurring, deconvolution, dehazing, matting, reflection removal and image interpolation, among others. With current image sizes in the order of megabytes, it is extremely exp ...

arXiv2018

Face Recognition in Real-world Images

Sabine Süsstrunk, Radhakrishna Achanta, Xavier Philippe Marie Fontaine

Face recognition systems are designed to handle well-aligned images captured under controlled situations. However real-world images present varying orientations, expressions, and illumination conditions. Traditional face recognition algorithms perform poor ...

Ieee2017