Stephan Brehm
YOU?
Author Swipe
View article: Controlling 3D Objects in 2D Image Synthesis
Controlling 3D Objects in 2D Image Synthesis Open
In this work, we propose a method that enforces explicit control over various attributes during the image generation process in a generative adversarial net. We propose a semi-supervised learning procedure that allows us to use a quantized…
View article: Semantically Consistent Image-to-Image Translation for Unsupervised Domain Adaptation
Semantically Consistent Image-to-Image Translation for Unsupervised Domain Adaptation Open
Unsupervised Domain Adaptation (UDA) aims to adapt models trained on a source domain to a new target domain where no labelled data is available. In this work, we investigate the problem of UDA from a synthetic computer-generated domain to …
View article: Semantically Consistent Image-to-Image Translation for Unsupervised Domain Adaptation
Semantically Consistent Image-to-Image Translation for Unsupervised Domain Adaptation Open
Unsupervised Domain Adaptation (UDA) aims to adapt models trained on a source domain to a new target domain where no labelled data is available. In this work, we investigate the problem of UDA from a synthetic computer-generated domain to …
View article: Semantically Consistent Image-to-Image Translation for Unsupervised\n Domain Adaptation
Semantically Consistent Image-to-Image Translation for Unsupervised\n Domain Adaptation Open
Unsupervised Domain Adaptation (UDA) aims to adapt models trained on a source\ndomain to a new target domain where no labelled data is available. In this\nwork, we investigate the problem of UDA from a synthetic computer-generated\ndomain …
View article: Semantic Consistency in Image-to-Image Translation for Unsupervised Domain Adaptation
Semantic Consistency in Image-to-Image Translation for Unsupervised Domain Adaptation Open
Unsupervised Domain Adaptation (UDA) aims to adapt models trained on a source domain to a new target domain where no labelled data is available. In this work, we investigate the problem of UDA from a synthetic computer-generated domain to …
View article: Error Bounds of Projection Models in Weakly Supervised 3D Human Pose Estimation
Error Bounds of Projection Models in Weakly Supervised 3D Human Pose Estimation Open
The current state-of-the-art in monocular 3D human pose estimation is heavily influenced by weakly supervised methods. These allow 2D labels to be used to learn effective 3D human pose recovery either directly from images or via 2D-to-3D p…
View article: Error Bounds of Projection Models in Weakly Supervised 3D Human Pose\n Estimation
Error Bounds of Projection Models in Weakly Supervised 3D Human Pose\n Estimation Open
The current state-of-the-art in monocular 3D human pose estimation is heavily\ninfluenced by weakly supervised methods. These allow 2D labels to be used to\nlearn effective 3D human pose recovery either directly from images or via\n2D-to-3…
View article: High-Resolution Dual-Stage Multi-Level Feature Aggregation for Single Image and Video Deblurring
High-Resolution Dual-Stage Multi-Level Feature Aggregation for Single Image and Video Deblurring Open
In this paper we address the problem of dynamic scene motion deblurring. We present a model that combines high-resolution processing with a multi-resolution feature aggregation method for single frame and video deblurring. Our proposed mod…
View article: A Convolutional Sequence to Sequence Model for Multimodal Dynamics Prediction in Ski Jumps
A Convolutional Sequence to Sequence Model for Multimodal Dynamics Prediction in Ski Jumps Open
A convolutional sequence to sequence model for predicting the jump forces of ski jumpers directly from pose estimates is presented. We collect the footage of multiple, unregistered cameras together with the output of force measurement plat…
View article: Multimodal Image Captioning for Marketing Analysis
Multimodal Image Captioning for Marketing Analysis Open
Automatically captioning images with natural language sentences is an\nimportant research topic. State of the art models are able to produce\nhuman-like sentences. These models typically describe the depicted scene as a\nwhole and do not t…
View article: A closer look: Small object detection in faster R-CNN
A closer look: Small object detection in faster R-CNN Open
Faster R-CNN is a well-known approach for object detection which combines the generation of region proposals and their classification into a single pipeline. In this paper we apply Faster R-CNN to the task of company logo detection. Motiva…