imagenet-1k-vl-enriched

Imagenet-1K-VL-Enriched在ImageNet-1K数据集的基础上进行了增强，提供了图像描述、边界框以及标签问题等信息。它包含超过128万张训练图像和5万张验证图像，并为每张图像提供了图像ID、原始标签，以及通过对象检测模型生成的边界框坐标、置信度分数和标签，以及BLIP2模型生成的图像描述。此外，它还标注了图像的质量问题，如重复、错误标记、过暗、模糊、过亮和异常值图像。该仓库遵循Apache 2.0许可协议，并提供了一个交互式可视化平台，以帮助用户更方便地浏览和分析数据。Imagenet-1K-VL-Enriched is an enhanced variant built upon the ImageNet-1K dataset, offering supplementary information including image captions, bounding boxes, and label-related issues. It contains over 1.28 million training images and 50,000 validation images. For each image, it provides the image ID, original label, bounding box coordinates, confidence scores and category labels generated by an object detection model, as well as image captions generated by the BLIP2 model. Additionally, it annotates various image quality issues, namely duplicate images, mislabeled samples, underexposed images, blurry images, overexposed images, and outlier images. This repository adheres to the Apache 2.0 open-source license, and offers an interactive visualization platform to facilitate users' browsing and data analysis.

搜集汇总

背景与挑战

背景概述

imagenet-1k-vl-enriched是ImageNet-1K数据集的增强版本，提供了图像描述、边界框和标签问题等额外信息，支持图像分类和目标检测等多种任务。数据集包含128万张训练图像和5万张验证图像，并遵循Apache 2.0许可协议。