datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
icir
i-CIR Dataset (Hugging Face)
website | arxiv | github
About
i-CIR (Instance-Level Composed Image Retrieval) is a curated benchmark for composed image retrieval where each instance corresponds to a specific, visually indistinguishable object (e.g., a particular landmark). Each query combines an image of the instance with a text modification, and retrieval is evaluated against a database containing rich hard negatives (visual / textual / compositional).
Key stats… See the full description on the dataset page: https://huggingface.co/datasets/billpsomas/icir.wallpapers-cog-iclThis is a web scrape of 7-themes.com, a wallpaper sharing website, along with metadata and captions from THUDM's CogVLM (vicuna-based).
Meta data is provided including the "file name" and the category in which each image was uploaded, which are used to provide contextual clues to THUDM's CogVLM to write captions for each image, leading to outstanding accuracy in captions, particular for the proper name of specific objects or characters.
Image files resized to a maximum of 2560x1440 (if over)… See the full description on the dataset page: https://huggingface.co/datasets/panopstor/wallpapers-cog-icl.TAMMs
TAMMs: Change Understanding and Forecasting in Satellite Image Time Series with a Temporal-Aware Multimodal Model
📄 Paper (ICLR 2026)
TAMMs is a large-scale dataset derived from the Functional Map of the World (fMoW) dataset, curated to support multimodal and temporal reasoning tasks such as change detection and future prediction.
Dataset Summary
This dataset contains 37,003 high-quality temporal sequences, each consisting of at least four distinct… See the full description on the dataset page: https://huggingface.co/datasets/IceInPot/TAMMs.Stability_Landscapes
📚 Overview: Kuramoto-Stability-Landscape (KSL)
The dataset consists of synthetic oscillator network topologies created using a random growth algorithm. Dynamical simulations are conducted by applying the second-order Kuramoto model to the nodes. This model is widely recognized for its effectiveness in analyzing synchronization dynamics in complex systems such as power grids and neuronal networks.
Two ensembles are included, each containing 10,000 unique network topologies:… See the full description on the dataset page: https://huggingface.co/datasets/PIK-ICoNe-landscape/Stability_Landscapes.icm-data-tempHAWK_ICCE2025
download
hf download backseollgi/HAWK_ICCE2025 --repo-type dataset --local-dir .
HAWK_bench 복원
1. 분할 파일 합치기
cat HAWK_bench.tar.gz.part-* > HAWK_bench.tar.gz
2. 압축 해제
tar -I pigz -xvf HAWK_bench.tar.gz
HAWK_bench_json 복원
1. 분할 파일 합치기
cat HAWK_bench_json.tar.gz.part-* > HAWK_bench_json.tar.gz
2. 압축 해제
tar -I pigz -xvf HAWK_bench_json.tar.gz
testICDAR-13-Logical
ICDAR-2013-Logical: A Line-Level Logical Conversion of the ICDAR 2013 Table Dataset
Dataset Description
This dataset is a converted and enhanced version of the ICDAR 2013 Table Competition dataset, specifically reformatted for modern Table Structure Recognition (TSR) and OCR tasks. 📜
The primary contribution of this version is the creation of a direct link between low-level OCR output and the table's logical structure. For each table, the dataset provides:
A… See the full description on the dataset page: https://huggingface.co/datasets/saeed11b95/ICDAR-13-Logical.ich-16icl_bimanual_dataICASSP2024-Acoustic_Scattering_AI-Noninvasive_Object_Classificationsicassp_audiococo10kMultilingualLibriMix