Forschung - Eliza

Startseite
Forschung

Hyperparameter Transfer Across Developer Adjustments

von Danny Stoll, Jörg K.H. Franke, Diane Wagner, Simon Selg, Frank Hutter

DOI: https://doi.org/10.48550/arXiv.2010.13117

This paper introduces a new research framework called hyperparameter transfer across adjustments (HT-AA) to address the challenge of automatically leveraging results from previous hyperparameter optimization (HPO) to accelerate new HPO after developer adjustments to machine learning algorithms. Developer adjustments can change which hyperparameter settings perform well or even modify the hyperparameter search space itself. The work provides four simple HT-AA baseline algorithms and eight benchmarks that change various aspects of ML algorithms, their hyperparameter search spaces, and neural architectures. The best baseline achieves 1.2-2.6x faster performance compared to prominent HPO algorithms without transfer, leading to faster development cycles, lower costs, and reduced environmental impacts.
ControlNet-XS: Rethinking the Control of Text-to-Image Diffusion Models as Feedback-Control Systems

von Denis Zavadski, Johann-Friedrich Feiden , Carsten Rother

DOI: https://doi.org/10.48550/arXiv.2312.06573

ControlNet-XS improves text-to-image generation by redesigning the communication between controlling networks and generation processes using feedback-control systems theory. The key innovation replaces sparse, low-bandwidth feedback signals with high-frequency, large-bandwidth communication, reducing delays that harm control system performance. This approach achieves better image quality and control fidelity while using significantly fewer parameters, making the model approximately twice as fast during inference and training. ControlNet-XS outperforms state-of-the-art methods for pixel-level guidance (depth, edges, segmentation) and matches performance for human pose guidance.
BIM-based AI-supported LiDAR-Camera Pose Refinement

von Miguel Arturo Vega Torres, Anna Ribic, Borja García de Soto, André Borrmann

DOI: https://doi.org/10.48550/arXiv.2412.03434

This paper introduces BIMCaP, a novel method to integrate mobile 3D sparse LiDAR data and camera measurements with pre-existing building information models (BIMs), enhancing fast and accurate indoor mapping with affordable sensors. BIMCaP refines sensor poses by leveraging a 3D BIM and employing a bundle adjustment technique to align real-world measurements with the model. Experiments using real-world open-access data show that BIMCaP achieves superior accuracy, reducing translational error by over 4 cm compared to current state-of-the-art methods. This advancement enhances the accuracy and cost-effectiveness of 3D mapping methodologies like SLAM. BIMCaP's improvements benefit various fields, including construction site management and emergency response, by providing up-to-date, aligned digital maps for better decision-making and productivity.

This site is registered on wpml.org as a development site. Switch to a production site key to remove this banner.