The study of inter-annotator agreement in musical pattern annotations has gained increased attention over the past few years. While expert annotations are often taken as the reference for evaluating pattern discovery algorithms, relying on just one reference is not usually sufficient to capture the complex musical relations between patterns. In this paper, we address the potential of digital annotation tools to enable large-scale annotations of musical patterns, by comparing datasets gathered with two recently developed digital tools. We investigate the influence of the tools and different annotator backgrounds on the annotation process by performing inter-annotator agreement analysis and feature-based analysis on the annotated patterns. We discuss implications for further adaptation of annotation tools, and the potential for deriving reference data from such rich annotation datasets for the evaluation of automatic pattern discovery algorithms in the future.

The task of reconstructing 3D scenes based on visual data represents a longstanding problem in computer vision. Common reconstruction approaches rely on the use of multiple volumetric primitives to describe complex objects. Superquadrics (a class of volumetric primitives) have shown great promise due to their ability to describe various shapes with only a few parameters. Recent research has shown that deep learning methods can be used to accurately reconstruct random superquadrics from both 3D point cloud data and simple depth images. In this paper, we extended these reconstruction methods to intensity and color images. Specifically, we used a dedicated convolutional neural network (CNN) model to reconstruct a single superquadric from the given input image. We analyzed the results in a qualitative and quantitative manner, by visualizing reconstructed superquadrics as well as observing error and accuracy distributions of predictions. We showed that a CNN model designed around a simple ResNet backbone can be used to accurately reconstruct superquadrics from images containing one object, but only if one of the spatial parameters is fixed or if it can be determined from other image characteristics, e.g., shadows. Furthermore, we experimented with images of increasing complexity, for example, by adding textures, and observed that the results degraded only slightly. In addition, we show that our model outperforms the current state-of-the-art method on the studied task. Our final result is a highly accurate superquadric reconstruction model, which can also reconstruct superquadrics from real images of simple objects, without additional training.

Facial composites, or police sketches, are essential tools in law enforcement for reconstructing the appearance of suspects from eyewitness descriptions. Traditionally, forensic artists manually produce these composites through intensive and time-consuming collaboration with witnesses. To improve the efficiency of this process, automated approaches have leveraged advancements in deep learning and generative modeling. However, the application of recent diffusion models has remained unexplored, despite their unparalleled text-guided synthesis capabilities. To this end, we present Diff-FIT (Diffusion Facial Identification Technique), a novel multi-pipeline framework for generating photorealistic facial composites in only a few steps with pretrained diffusion models. Diff-FIT enables rapid generation of initial composites from textual descriptions, followed by intuitive sequential edits including global image-to-image translation, local text-based inpainting, and drag-based geometric transformations. Through experiments across multiple latent diffusion models and sampling parameters we determine the configuration that best balances image quality, diversity, image-text alignment, and identity consistency. In a user study involving biometric experts and non-experts, Diff-FIT achieves comparable real-world utility to state-of-the-art systems in both subjective evaluations and identification rates with generated facial composites, while enabling greater variation and flexibility through description-based generation and diverse editing pipelines for adding distinct facial features. The source code for the Diff-FIT framework is publicly available at: https://github.com/matemato/Diff-FIT.

Current state-of-the-art segmentation techniques for ocular images are critically dependent on large-scale annotated datasets, which are labor-intensive to gather and often raise privacy concerns. In this paper, we present a novel framework, called BiOcularGAN, capable of generating synthetic large-scale datasets of photorealistic (visible light and near-infrared) ocular images, together with corresponding segmentation labels to address these issues. At its core, the framework relies on a novel Dual-Branch StyleGAN2 (DB-StyleGAN2) model that facilitates bimodal image generation, and a Semantic Mask Generator (SMG) component that produces semantic annotations by exploiting latent features of the DB-StyleGAN2 model. We evaluate BiOcularGAN through extensive experiments across five diverse ocular datasets and analyze the effects of bimodal data generation on image quality and the produced annotations. Our experimental results show that BiOcularGAN is able to produce high-quality matching bimodal images and annotations (with minimal manual intervention) that can be used to train highly competitive (deep) segmentation models (in a privacy aware-manner) that perform well across multiple real-world datasets. The source code for the BiOcularGAN framework is publicly available at https://github.com/dariant/BiOcularGAN.