While mobile devices have developed into hardware with advanced capabilities for rendering 3D graphics, they commonly lack the computational power to render large 3D scenes with complex lighting interactively.A prominent approach to tackle this is rendering required views on a remote server and streaming them to the mobile client.However, the rate at which servers can supply data is limited, e.g., by the available network speed, requiring imagebased rendering techniques like image warping to compensate for the latency and allow a smooth user experience, especially in scenes where rapid user movement is essential.In this paper, we present a novel streaming approach designed to minimize artifacts during the warping process by including an additional visibility layer that keeps track of occluded surfaces while allowing access to 360 views.In addition, we propose a novel mesh generation technique based on the detection of loops to reliably create a mesh that encodes the depth information required for the image warping process.We demonstrate our approach in a number of complex scenes and compare it against existing works using two layers and one layer alone.The results indicate a significant reduction in computation time while achieving comparable or even better visual results when using our dual-layer approach.

Multilevel image thresholding is a simple and efficient segmentation technique.Thresholding criteria such as Otsu and Kapur objective functions are extensively used in the literature.They are effective techniques but suffer from poor computational complexity.Thus, methods such as dynamic programming for exact optimization or metaheuristic algorithms for approximate optimization are applied to improve runtime.However, most of these algorithms take the count of thresholds as input.Hence, a novel input-less algorithm that can identify the count and values of thresholds simultaneously is proposed.The proposed method is then compared to state of the art methods to assess its efficiency and effectiveness.

In recent years, the demand for realistic human avatars has escalated across diverse industries, ranging from gaming and virtual or augmented reality to fashion and healthcare.Creating an accurate, lifelike virtual duplicate of a human subject requires a precise reconstruction of anthropometric measurements from the real body to the rendered body model.In this paper, we propose a novel pipeline for 3D human body model reconstruction from a set of two input images.Our method generates metrically accurate virtual human avatars, based on body measurements extracted from an input image.Our approach is capable of parameterizing the generated mesh with body measurements and yields accurate results.To augment the metrically accurate body meshes, we introduce a pipeline for generating textured clothes to enhance user virtual experience.As part of our work, we generated a synthetic dataset that serves as a foundation for further enhancing the training process by extracting anthropometric measurements from an input photo.This dataset contains over one million files derived from 120,000 virtual avatars that can extend existing real datasets.

Accurate recognition and interpretation of sign language are crucial for enhancing communication accessibility for deaf and hard of hearing individuals.However, current approaches of Isolated Sign Language Recognition (ISLR) often face challenges such as low data quality and variability in gesturing speed.This paper introduces a comprehensive model training pipeline for ISLR designed to accommodate the distinctive characteristics and constraints of the Sign Language (SL) domain.The constructed pipeline incorporates carefully selected image and video augmentations to tackle the challenges of low data quality and varying sign speeds.Including an additional regression head combined with IoU-balanced classification loss enhances the model's awareness of the gesture and simplifies capturing temporal information.Extensive experiments demonstrate that the developed training pipeline easily adapts to different datasets and architectures.Additionally, the ablation study shows that each proposed component expands the potential to consider ISLR task specifics.The presented strategies enhance recognition performance across various ISLR benchmarks and achieve state-of-the-art results on the WLASL and Slovo datasets.

Computer-Aided Design (CAD) has revolutionized design and manufacturing by enabling precise, complex models in collaborative environments.While similar CAD models with application-specific modifications are often required, designs are typically created from scratch due to challenges in retrieving existing models or generating editable ones.Although parametric CAD modeling has advanced through deep generative approaches treating CAD as a language task to generate user-editable designs, building truly scalable multi-modal datasets and networks tailored for 3D design tasks, particularly in engineering domains remains a significant challenge.Developing such datasets, especially those incorporating images, point clouds and user-like text and hand-drawn sketches is difficult as these modalities demand fine-grained geometric understanding and extensive human-in-the-loop evaluations.While large foundational models like CLIP have improved cross-modal retrieval, they are primarily trained on natural images and fail to capture the geometric and structural complexities inherent to CAD data.In this paper, we propose a novel multi-modal pipeline for CAD command sequence generation using state-of-theart Vision-Language Models (VLMs).We introduce a unique multimodal CAD dataset comprising hand-drawn sketches, CAD command sequences, images and basic text prompts.These modalities are integrated through a Multi-modal Retrieval-Augmented Generation (MM-RAG) framework to enable user-editable CAD model retrieval and generation.Our RAG-based pipeline streamlines the CAD design process by enabling iterative, userguided model generation based on simple sketches or text queries.This approach aims to streamline CAD model design by creating an advanced, end-to-end pipeline that supports design workflows.The dataset and code will be made publicly available at: https://github.com/ananthu2014/cadrag.

Ensuring the correctness of annotations in training datasets is one way to increase the trustworthiness and reliability of Machine Learning.This study aims to detect semantic shifts in datasets using Feature-Based Out-Of-Distribution and outlier detection methods, assuming Out-Of-Distribution samples are far from In-Distribution data.The experiments began with distance-based methods, such as k-Nearest Neighbours and Mahalanobis, followed by feature pyramids and dimensionality reduction techniques to address high-dimensional challenges.The results showed that the k-Nearest Neighbours detector performed robustly, achieving 100% AUROC when using ResNet50 on the Caltech-101 dataset, while the Mahalanobis detector showed unstable results with scores close to 50%.Moreover, selecting the right backbone model and feature levels, particularly low-level features from ResNet50, improved performance achieving AUROC score of 96% on the DelftBikes dataset for both k-Nearest Neighbours and Local Outlier Factor.The study highlights that k-Nearest Neighbours, Local Outlier Factor, alongside feature pyramids and dimensionality reduction constitute an effective setup for Out-of-Distribution detection, but optimal performance depends on tailored configurations across varying data conditions.

With growing General Data Protection Regulation (GDPR) compliance demands for deep learning surveillance models, human anonymization is a key research area. Most studies use RGB images as input for generative models, which retain demographic features, compromising anonymization and consistency across frames. We present our initial study into a full-body anonymization pipeline for anomaly detection datasets, where the synthetic person generation never has access to the RGB pedestrian visuals. The proposed pipeline uses a combination of existing models for easier reproducibility. We use YoloV8 for object detection, ClipSeg and BiRefNet for segmentation, OpenPose for pose estimation, and an animation diffusion model. The diffusion model processes only masks and skeletal pose images, removing the problems with using sensitive data. We test on the Avenue dataset. We show that the proposed pipeline can consistently anonymize and change the demographics of detected pedestrians. We discuss the observed problems and the next steps in building a more robust second version.