Abstract Efficient data ingestion and online data augmentation remain challenges in deep learning workflows, particularly when dealing with datasets containing non-standard formats or massive multidimensional arrays that natively optimised functions cannot fully manage. This work presents a parallel framework that integrates and global shared memory through a ring buffer architecture, enabling high-throughput data loading and flexible on-the-fly augmentation. The framework decouples data production from consumption, allowing multiple CPU workers to load and preprocess batches in parallel while completely bypassing the Python GIL and memory bottlenecks. Crucially, the framework supports both CPU-side and GPU-side augmentation strategies, adapting to whether complex conditional transformations or framework-native operations are required. The proposed approach was validated on two representative tasks: (i) sign language recognition from human pose CSV sequences, and (ii) hyperspectral image classification using massive arrays. Relative to standard sequential baselines, the proposed framework achieved up to $$27\times $$ 27 × acceleration in isolated data ingestion and up to $$28\times $$ 28 × in end-to-end training. Importantly, even against natively optimised parallel TensorFlow and PyTorch pipelines, it still delivered up to $$8\times $$ 8 × faster data loading and up to $$7\times $$ 7 × faster full training in memory-intensive scenarios. Overall, the proposed framework provides a scalable, multi-GPU compatible solution for deep learning pipelines, showing robust performance across both I/O-bound and memory-constrained scenarios in TensorFlow and PyTorch while alleviating memory fragmentation and allocation constraints.

Nowadays, smart environments (SEs) enable the monitoring of people with physical disabilities by incorporating activity recognition. Thermal cameras are being incorporated as they preserve privacy. Some deep learning (DL) solutions use the pose of the users because it removes external noise. Although there are robust DL solutions in the visible spectrum (VS), they fail in the thermal domain. Thus, we propose thermal human pose lite (THPoseLite), a convolutional neural network (CNN) based on MobileNetV2 that extracts pose from thermal images (TIs). In a novel way, an auto-labeling approach has been developed. It includes a background removal using an optical flow estimator. It also integrates Blazepose [a pose estimator for VS images (VSIs)] to obtain the poses in the preprocessed TIs. Results show that the preprocessing increases the percentage of detected poses by Blazepose from 19.55% to 76.85%. This allows the recording of human pose estimation (HPE) data sets in the VS without requiring VS cameras or manually annotating data sets. Furthermore, THPoseLite has been embedded in an Internet of Things (IoT) device incorporating an edge tensor processing unit (TPU) accelerator, which can process TIs recorded at 9 frames per second (FPS) in real time (12.28 FPS). It requires fewer than 6W of energy to run. It has been achieved using model quantization, decreasing the accuracy in estimating the poses by only 1%. The mean-squared error of MobileNetV2 in test images is 35.48, obtaining accurate poses in 21% of the images that Blazepose is not able to detect any pose.

Abstract The intrinsic high parallelism and entanglement characteristics of quantum computing have made quantum image processing techniques a focus of great interest. One of the most widely used techniques in image processing is segmentation, which in one of their most basic forms can be carried out using thresholding algorithms. In this paper, a fault-tolerant quantum dual-threshold algorithm has been proposed. This algorithm has been built using only Clifford+T gates for compatibility with error detection and correction codes. Because fault-tolerant implementation of T gates has a much higher cost than other quantum gates, our focus has been on reducing the number of these gates. This has allowed adding noise tolerance, computational cost reduction, and fault tolerance to the state-of-the-art dual-threshold segmentation circuits. Since the dual-threshold image segmentation involves the comparison operation, as part of this work we have implemented two full comparator circuits. These circuits optimize the metrics T-count and T-depth with respect to the best circuit comparators currently available in the literature.

This is a work focused on remote browsing of JPEG2000 image sequences which takes advantage of the spatial scalability of JPEG2000 to determine which precincts of a subsequent image should be transmitted, and which precincts should be reused from a previously reconstructed image. The results of our experiments demonstrate that the quality of the reconstructed images can be significantly increased by using motion compensation and conditional replenishment on the client side. The proposed algorithm is compatible with standard JPIP servers.

This work proposes the video codec MCJ2K (Motion Compensated JPEG2000), which is based on Motion Compensated Temporal Filtering (MCTF) and JPEG2000 (J2K). MCJ2K exploits the temporal redundancy present in most videos, thereby increasing the rate/distortion performance, and generates a collection of temporal subbands which are compressed with J2K. MCJ2K code-streams can be managed by standard JPIP (J2K Interactive Protocol) servers.

Fully connected push-based P2P overlay networks, such as P2PSP (P2P Straightforward Protocol), are an efficient alternative to tree-shaped overlay ones. In P2PSP, as in many other P2P systems, malicious peers (MPs) can perform collaborative pollution and free-riding attacks. Here, we summarize current solutions for these problems. We are not interested in make the content private but in mitigating the previous attacks. The proposed solution is based on Shamir's Secret Sharing (SSS) and the use of trusted peers (TPs). Under the assumption that the number of malicious peers is smaller than the half of the total, our proposal forces any peer to relay unpolluted content to well-intended peers in order to not to be expelled. We show the overhead added to the protocol, its strengths, its weaknesses, and outline possible solutions for these weakness.

The JPEG2000 image coding system was created with the intention of superseding the original JPEG standard, using a novel wavelet-based method. The main advantage of JPEG2000 is the flexibility of its code-stream, which provides new functionality related to the interactive transmission of images. For this task, JPEG2000 uses the JPIP protocol, which enables real-time spatial random access while the retrieved image is progressively displayed (streaming). The standard also foresees the compression and transmission of sequences of images by repeating this approach for each image. In this framework, this paper presents the Continue data-flow control strategy, a JPIP-compliant solution for the interactive streaming of sequences of images that are transmitted over time-varying communication channels. In this context, the random fluctuation of the capacity of the transmission channel over the time forces the clients to prefetch a minimal amount of the code-stream of each image of the beginning of the transmitted sequence before the playback starts, and the server to decide, in real-time, which amount of the code-stream of each compressed image is going to be transmitted . The estimated channel capacity is performed by clients and the rate-control at the server is straightforward, resulting in a highly scalable image retrieval system. The experiments conducted in this study demonstrate that the proposed method keeps a constant playback frame-rate under severe variations of the channel capacity, even when short prefetch times are used.

Hyperspectral data lossy compression has not yet achieved global acceptance in the remote sensing community, mainly because it is generally perceived that using compressed images may affect the results of posterior processing stages. This possible negative effect, however, has not been accurately characterized so far. In this letter, we quantify the impact of lossy compression on two standard approaches for hyperspectral data exploitation: spectral unmixing, and supervised classification using support vector machines. Our experimental assessment reveals that different stages of the linear spectral unmixing chain exhibit different sensitivities to lossy data compression. We have also observed that, for certain compression techniques, a higher compression ratio may lead to more accurate classification results. Even though these results may seem counterintuitive, this work explains these observations in light of the spatial regularization and/or whitening that most compression techniques perform and further provides recommendations on best practices when applying lossy compression prior to hyperspectral data classification and/or unmixing.

This papers studies a novel prefetching scheme for the remote browsing of sequences of high resolution JPEG 2000 images. Using this scheme, an user is able to select randomly any of the remote images for its analysis, repeating this process with other images after some undefined time. Our solution has been proposed in a low bit-rate communication context where the complete transmission of any of the images for its lossless recovery should take too much time for an interactive visualization. For this reason, quality scalability is used in order to minimize the decoding latency. Frequently, the user can also play a "video'', moving sequentially on the neighbour (consecutive in time over previous or following) images of the currently displayed one. With the objective of hiding also the link latency, the proposed data scheduler transmits in parallel data of the image that it is currently displayed and data of the rest of the temporally adjacent images. This scheduler uses a model based on the quality progression of the image in order to estimate which percentage of the bandwidth is dedicated to prefetch data. Our experimental results prove that a significant benefit can be achieved in terms of both subjective quality and responsiveness by means of prefetching.

This paper describes and analyzes JPIP-W, an innovative proposal for the interactive transmission of JPEG2000 images on the Internet. JPIP-W is an extension of JPIP, the current JPEG protocol proposed for interactive JPEG2000 image browsing. One of the JPIP characteristics of greatest interest is its ability to use the Web for retrieving images. However, JPIP is unable to exploit the large infrastructure of today's Web caching systems (proxies), used to reduce response time and network traffic. To overcome this drawback, JPIP-W defines a new server-client interaction consisting of splitting JPEG2000 images into data blocks that can be cached by the proxies. These blocks can be shared among several clients, allowing fast recovery of some portions of the images. Experimental results demonstrate that JPIP-W significantly reduces JPIP retrieving times.

This paper studies a progressive image transmission technique over waveform channels. The channel optimized vector quantization codec (COVQ) (Farvardin and Vaishampayan 1991) is applied to the image wavelet coefficients creating a robust progressive image transmission technique that mitigates the effects of a noisy channel on the reconstructed image. In order to evaluate the performance of our proposal, a Gaussian and slow-fading Rayleigh channel model, with several different values of channel signal to noise ratio (CSNR) were simulated in our experiments. Examples show a significant visual improvement of our application compared to other progressive image transmission techniques.