open source computer vision (opencv)

To address the high computational overhead and memory bottlenecks hindering Vision Transformer (ViT) deployment on edge devices, this paper presents a systematic survey of lightweighting and acceleration techniques tailored for edge scenarios—spanning model compression (e.g., pruning, quantization, knowledge distillation, attention simplification), software optimization (e.g., compiler frameworks such as TVM), and hardware adaptation (e.g., GPU/TPU/FPGA mapping). Its key contributions include: (1) proposing the first unified taxonomy for ViT edge deployment, explicitly characterizing trade-offs among accuracy, latency, power consumption, and hardware platforms; (2) establishing a structured evaluation framework covering 120+ works to identify real-world deployment bottlenecks; and (3) delivering a reproducible, cross-platform technical selection guide to advance co-optimization of accuracy, latency, and power efficiency.

This work challenges the prevailing notion that handcrafted local features such as SIFT should be superseded by learned methods, addressing the lack of a unified GPU-based framework for fair and modular comparison between the two paradigms. The authors introduce PySIFT—the first fully GPU-resident, bit-wise deterministic implementation of SIFT—built on CuPy and Numba, and seamlessly integrated with major deep learning frameworks via DLPack for zero-copy interoperability. Experiments demonstrate that classical features and learned matchers are complementary rather than mutually exclusive. On an RTX 3050, PySIFT outperforms OpenCV’s SIFT in both speed and accuracy: it achieves higher MMA on HPatches, reduces per-image-pair processing time on MegaDepth by 383 ms, and significantly improves cross-dataset geometric accuracy (e.g., +5.6 percentage points in AUC@10°), while ensuring consistent outputs across GPU architectures—overcoming the non-determinism inherent in cuDNN.

JPEG decoding bottlenecks in machine learning vision pipelines severely impede I/O throughput. To address this, this work presents the first systematic cross-platform evaluation of nine mainstream Python JPEG decoders—including libjpeg-turbo, Pillow, OpenCV, and PyTorch—across throughput, latency, and memory footprint on ARM64 (Apple M4 Max) and x86_64 (AMD Threadripper) architectures. Leveraging empirical benchmarking under realistic training and inference workloads, we derive a principled, architecture-aware decoder selection guideline. Our results show that libjpeg-turbo delivers an average 1.5× speedup over alternatives while maintaining low memory overhead and consistent latency characteristics. The study provides reproducible, production-ready optimization insights, directly reducing image loading overhead in both distributed training and real-time inference scenarios. All benchmarks, configurations, and analysis scripts are publicly released to ensure transparency and deployability.

The absence of a unified evaluation benchmark for video compression optimized for downstream vision tasks (e.g., detection, recognition) hinders standardization of Feature Coding for Machines (FCM). Method: This paper introduces the first open-source evaluation platform supporting joint optimization across multiple dimensions: vision tasks (image/video understanding), model architectures (CNNs/Transformers), and codecs (traditional and deep learning-based). It proposes a novel rate-distortion–task-accuracy joint analysis framework, integrating standardized codecs with differentiable neural compression modules to enable end-to-end co-evaluation of compression efficiency and downstream task performance. Contribution/Results: Extensive validation on multiple benchmark datasets demonstrates significant improvements in task-driven compression efficiency. The platform has been formally adopted by MPEG as the core infrastructure for FCM standard development and evaluation.

Existing image annotation tools require pre-uploading datasets, making them ill-suited for real-time data acquisition from experimental instruments such as microscopes—thus hindering rapid AI model iteration in scientific research. To address this, we propose a real-time image annotation system tailored for scientific experimentation. It integrates camera video streams via OpenCV and NumPy, enabling low-latency frame capture and interactive annotation within a Python-based graphical user interface. The system introduces three key innovations: Bezier spline-based contour drawing, editable binary masks, and non-destructive layer management—supporting on-demand, high-precision, and fully traceable annotations. Compatible with diverse video input devices, it unifies data acquisition, annotation, and model training into a seamless closed-loop workflow. This significantly accelerates AI model development and deployment in experimental settings. The implementation is open-sourced.

This study addresses the limitations of conventional image enhancement, filtering, and pattern recognition—namely, heavy reliance on manual feature engineering and insufficient real-time performance—by proposing a theory-driven, end-to-end machine learning framework. Methodologically, it is the first to systematically integrate discrete Fourier transform (DFT), Z-transform, and continuous Fourier analysis into deep learning pipelines, synergistically coupling them with convolutional neural networks (CNNs) and classical digital filtering algorithms to enable frequency-domain-guided automated feature extraction and real-time joint signal–image processing. The key contributions include: (i) development of an extensible Python framework; (ii) average PSNR improvement of 3.2 dB in image enhancement and noise suppression tasks; and (iii) 40% acceleration in feature extraction efficiency. This work establishes a novel paradigm for AI-powered real-time computer vision that simultaneously ensures high performance and interpretability.

This paper addresses three critical challenges in high-resolution image processing: grayscale quantization distortion, low feature extraction accuracy, and non-invertible transformations. To this end, we propose a modular spatial image processing framework. Methodologically, it employs a multi-level collaborative algorithm architecture integrating structure-preserving 8-level grayscale discretization, RGB/YCrCb histogram equalization, HSV brightness adaptive adjustment, 3×3 convolution-based sharpening and unsharp masking, gamma correction, and joint extraction of Canny edges, Hough lines, Harris corners, and morphological geometric features. Additionally, a reversible bidirectional transformation pipeline is designed to ensure consistency between forward processing and inverse reconstruction. Experimental results demonstrate a bidirectional transformation fidelity of 76.10% (forward) and 74.80% (inverse), cue stick angle estimation error below 0.5° (measured at 51.50°), cue isolation similarity to the original image of 81.87%, and strong robustness across multiple benchmark datasets.

This study addresses the insufficient robustness and accuracy of local feature matching in overlapping regions of satellite imagery. To this end, the authors construct a manually curated satellite image dataset annotated with GPS coordinates and conduct a systematic evaluation of SIFT and ORB algorithms across the entire matching pipeline—including keypoint detection, descriptor extraction, feature matching, and RANSAC-based geometric verification. Using the inlier ratio as the primary metric for matching quality, the work quantitatively analyzes the impact of keypoint quantity on matching performance. The results reveal a nonlinear relationship between the number of detected keypoints and the inlier ratio, offering empirical evidence and theoretical guidance for algorithm selection and parameter tuning in remote sensing image matching tasks.

This study systematically evaluates feature extraction methods for image recognition, focusing on Vision Transformers (ViTs) versus traditional handcrafted features (SIFT/SURF/ORB), CNNs, GANs, and deep feature models. Methodologically, we establish a unified experimental framework comparing both non-contrastive and contrastive learning paradigms, incorporating patch embedding, multi-head self-attention, positional encoding, and self-supervised strategies for feature modeling. Our key contributions are: (i) the first empirical demonstration across multiple benchmarks that ViTs—leveraging global contextual modeling—significantly outperform CNNs and handcrafted features in generalization, few-shot adaptation, and cross-domain transfer; and (ii) evidence that contrastive learning further enhances the discriminative power of ViT-derived features. The results provide a reproducible benchmark analysis and principled guidance for architecture selection in visual representation learning.

Although the DCVC family of deep video codecs achieves state-of-the-art rate-distortion performance, its publicly available code supports only inference—severely hindering reproducibility, benchmarking, and algorithmic advancement. Method: We present the first open-source, PyTorch-based deep video compression framework that unifies end-to-end training, validation, and testing for four representative DCVC models. The framework integrates temporal modeling with hybrid entropy coding and introduces context-adaptive probability modeling to enhance compression efficiency. Contribution/Results: Comprehensive benchmarking across standard datasets demonstrates average BD-rate reductions of 15.2%–28.7% relative to HEVC. This work fills a critical gap by providing the first trainable, open-source DCVC implementation, establishing a transparent, extensible, and fully reproducible infrastructure for learning-based video coding research.

This work addresses the challenge that existing feature extraction methods struggle to simultaneously achieve computational efficiency, accuracy, and cross-domain consistency when processing terabyte- to petabyte-scale 2D/3D image data. To this end, we propose Nyxus—a high-performance feature extraction framework that supports out-of-core computation and CPU/GPU parallel acceleration, featuring a unified architecture for programmable feature optimization and multi-platform deployment. Nyxus offers diverse interfaces including a Python package, command-line tools, a Napari plugin, and an OCI container, making it broadly applicable to biomedical domains such as radiomics and cellular analysis. Benchmark evaluations demonstrate that Nyxus significantly enhances the efficiency and scalability of large-scale AI workflows in both cloud and high-performance computing environments.