The article is devoted to a practical way to adapt image-based detector models, such as YOLO and RF-DETR, to video without changing their basic architecture. The approach is based on the formation of a superframe - a three-channel image from adjacent grayscale frames, where the channels encode temporal context instead …

Lightweight and fast neural network for tracking volleyballs: 200+ FPS on CPU (Intel i5-10400F + OpenVINO) Result: vball-net’s own model (variants V1 / FastV1 / V2) produces 200+ FPS on an Intel i5-10400F CPU in inference (OpenVINO), while maintaining decent tracking quality even on amateur recordings from Action Cam. → …

Volleyball is a popular sport in Indonesia and is widely taught in educational institutions. Due to the COVID-19 pandemic, training was transferred to an online format. One of the tools for distance learning is the website. The purpose of the study was to determine the effectiveness of teaching passing techniques …

The article discusses the problem of automatic analysis of video recordings of volleyball matches with an emphasis on identifying key points of the volleyball court and net. A comparative analysis of various neural network architectures (U-net, TrackNet, YOLO11-pose) for solving the problem of detecting key points on a sports ground …

FastSAM is an ultra-fast alternative to SAM for object segmentation. Based on YOLOv8-seg + YOLACT, uses CNN instead of ViT - 50 times faster with comparable quality. Supports hints: dots, boxes, text (via CLIP). Trained on 2% SA-1B data. Ideal for real-time applications.

Track a volleyball in real time at an incredible 200 FPS on a regular CPU! The lightweight ONNX model stores ball coordinates in CSV and produces a rendered video. Ideal for sports analytics and computer vision research. Try it on GitHub!

This paper examines the problem of accurate and efficient detection of a volleyball in real-time video using sequences of grayscale frames. A modification of the TrackNetV4 architecture is proposed with the introduction of spatial and temporal attention mechanisms to improve the localization of a small fast moving object. The model …

DeepBall is a specialized neural network model for ball detection in sports videos, designed to process images of any size and create a confidence map of the ball's position. It is based on a fully convolutional architecture and uses the concept of hypercolumns, combining feature maps from different levels of …