ShrimpView: A Versatile Dataset for Shrimp Detection and Recognition

The "ShrimpView: A Versatile Dataset for Shrimp Detection and Recognition" is a meticulously curated collection of 10,000 samples (each with 11 attributes) designed to facilitate the training of deep learning models for shrimp detection and classification. Each sample in this dataset is associated with an image and accompanied by 11 categorical attributes. These attributes span a range of features including species type ("Pacific White Shrimp," "Tiger Prawn," "Ghost Shrimp"), life stage ("Larvae," "Juvenile," "Adult"), and environmental conditions ("Freshwater," "Saltwater," "Brackish Water"), among others. Additionally, the dataset incorporates variations in image orientation, background, and lighting conditions to enhance model generalizability. With resolutions ranging from 640x480 px to 1920x1080 px, the dataset is well-suited for both object detection and multi-class classification tasks. The variability in these attributes aims to improve the model's generalizability and robustness. For instance, the "Species" and "Life Stage" attributes can aid in multi-class classification, while "Color" and "Size" add complexity for object detection. "Orientation" and "Background" introduce viewpoint and environmental variance, respectively. "Lighting" conditions can simulate different capture scenarios, and "Resolution" offers scale variance. "Grouping" prepares the model for detecting multiple instances in a single frame, and "Habitat" ensures the model is trained for different water conditions. It aims to serve as a robust foundation for developing comprehensive machine learning solutions in the domain of aquatic species recognition. This rich, multi-dimensional dataset is thus ideal for training a robust deep learning model for comprehensive shrimp detection and classification. The computational costs involved in generating the dataset are listed as follows:Data Augmentation Cost: Caug=20 secondsCSV Generation Cost: Ccsv=1.1 secondsAttribute Randomization Cost: Cattr=1.1 secondsTotal Disk Space Needed: Dtotal=550 MBTotal Time Needed: CTotal=22.2 secondsTotal Time Needed with Parallelization: Ctotal, Cparallel=55.5 seconds (4 cores)

搜集汇总

背景与挑战

背景概述

该数据集是一个包含10,000个样本的虾类检测和识别数据集，每个样本包含一张图片和11个分类属性，适用于深度学习模型的训练。数据集涵盖了多种虾类、生命阶段和环境条件，旨在提高模型的泛化能力和鲁棒性。