Back to datasets
Dataset assetOpen Source CommunityMachine LearningHuman Feedback
TIFA, AIGCIQA2023, ECCV_Caption, D3PO
This repository contains multiple datasets, including TIFA, AIGCIQA2023, ECCV_Caption, and D3PO, for human feedback research on reward models. Each dataset comes with detailed download and setup instructions.
Source
github
Created
Jan 5, 2024
Updated
Feb 5, 2024
Signals
193 views
Availability
Linked source ready
Overview
Dataset description and usage context
Dataset Overview
Dataset Download and Setup
Image Download
- TIFA and AIGCIQA2023 dataset images: Can be downloaded from this link. After downloading, extract to obtain the
annotated_imagesandallimgfolders.
TIFA Dataset
- Preference ranking data: located at
csvs/tifa.csv. - Image setup: Ensure the
annotated_imagesfolder is in the same directory astifa.csv.
AIGCIQA2023 Dataset
- Preference ranking data: located at
csvs/aigciqa2023.csv. - Image setup: Ensure the
allimgfolder is in the same directory asaigciqa2023.csv.
ECCV_Caption Dataset
- Preference ranking data: located at
eccv_caption.csv. - Image setup: Download the COCO2014 val split, extract, and place the
val2014folder in the same directory aseccv_caption.csv.
D3PO Dataset
- Image download: Can be downloaded from this link.
- Image setup: After extracting, place the
text2image_evaluation/imgfolder in the same directory as the CSV files of the D3PO dataset.
Need downstream help?
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.