SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing

Authors: Zhecheng Wang, Rajanie Prabha, Tianyuan Huang, Jiajun Wu, Ram Rajagopal

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental With continual pre-training on this dataset, we obtain a VLM that surpasses baseline models with a 6.2% average accuracy gain in zero-shot scene classification across seven benchmark datasets.
Researcher Affiliation Academia Stanford University {zhecheng, rajanie, tianyuah, ramr}@stanford.edu, jiajunwu@cs.stanford.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes 1The dataset and associated models are publicly available at https://github.com/wangzhecheng/Sky Script
Open Datasets Yes 1The dataset and associated models are publicly available at https://github.com/wangzhecheng/Sky Script
Dataset Splits No The paper sets aside 30,000 image-text pairs for testing cross-modal retrieval and mentions an auxiliary classification dataset, but it does not specify the training/validation/test splits for the main Sky Script dataset or how the auxiliary dataset is partitioned for validation and training.
Hardware Specification Yes The continual pre-training is conducted on 4 NVIDIA A100 GPUs with a batch size of 512 and total epochs of 20.
Software Dependencies No The paper does not provide specific software dependency versions (e.g., Python 3.8, PyTorch 1.9) for reproducibility.
Experiment Setup Yes The continual pre-training is conducted on 4 NVIDIA A100 GPUs with a batch size of 512 and total epochs of 20.