fassabilf/clip-pelatnas-p2-2026
收藏资源简介:
CLIP Pelatnas P2 2026 — ARIA Multimodal Crisis是一个多模态学习竞赛数据集,用于印度尼西亚IOAI 2026培训计划。数据集包含4个任务:零样本图像分类(使用STL-10数据集的200张测试图像和类别名称)、线性探测分类(使用STL-10的1000张标注训练图像和200张测试图像)、图像-文本检索(使用Flickr8k的6000个图像-标题对和200个查询)以及多选问答(MCQA,使用ScienceQA的2000个训练问题和200个测试问题)。总项目数为800个(每个任务200个),评估指标为准确率。数据集旨在模拟ARIA(AI研究站)因太阳耀斑损坏后重建的场景,用于测试CLIP模型在多模态任务中的性能,目标准确率超过85%。
CLIP Pelatnas P2 2026 — ARIA Multimodal Crisis is a multimodal learning competition dataset designed for Indonesia's IOAI 2026 training program. The dataset encompasses four tasks: 1) Zero-shot image classification, utilizing 200 test images and category names from the STL-10 dataset; 2) Linear probe classification, employing 1,000 annotated training images and 200 test images from the STL-10 dataset; 3) Image-text retrieval, using 6,000 image-caption pairs and 200 queries sourced from Flickr8k; 4) Multiple-Choice Question Answering (MCQA), adopting 2,000 training questions and 200 test questions from ScienceQA. The total number of samples is 800, with 200 samples per task, and accuracy is used as the evaluation metric. This dataset simulates the post-damage reconstruction scenario of ARIA (AI Research Station) caused by solar flares, aiming to test the performance of CLIP models on multimodal tasks, with a target accuracy exceeding 85%.




