datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
text-2-image-Rich-Human-Feedback
Building upon Google's research Rich Human Feedback for Text-to-Image Generation we have collected over 1.5 million responses from 152'684 individual humans using Rapidata via the Python API. Collection took roughly 5 days.
If you get value from this dataset and would like to see more in the future, please consider liking it.
Overview
We asked humans to evaluate AI-generated images in style, coherence and prompt alignment. For images that contained flaws, participants were… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/text-2-image-Rich-Human-Feedback.text-2-image-human-preferences-2m
Text-to-image human preferences: 2M votes across 30 models
This dataset contains the complete voting record behind the
Datapoint Image Bench
leaderboard: 2,161,160 validated pairwise votes — exactly 10 for each of
216,116 image pairs. The votes compare 30 text-to-image models in a complete
round-robin on 500 prompts, judged by annotators from over 200 countries.
Every vote includes the annotator's trust score at the time the vote was
cast.
Built on the Datapoint annotation… See the full description on the dataset page: https://huggingface.co/datasets/datapointai/text-2-image-human-preferences-2m.Textual-Image-Caption-Dataset
Update: OCT-2023
Add v2 with recent SoTA model swinV2 classifier for both soft/hard-label visual_caption_cosine_score_v2 with person label (0.2, 0.3 and 0.4)
Introduction
Modern image captaining relies heavily on extracting knowledge, from images such as objects,
to capture the concept of static story in the image. In this paper, we propose a textual visual context dataset
for captioning, where the publicly available dataset COCO caption (Lin et al., 2014) has been… See the full description on the dataset page: https://huggingface.co/datasets/AhmedSSabir/Textual-Image-Caption-Dataset.cc0-textures
Dataset Card for CC0 Textures
Dataset Summary
This dataset contains 18,785 texture images from cc0-textures.com. It includes textures of wood, metal, concrete, fabric, stone, ceramic, and other materials. The original archives were downloaded, unpacked, and images were compressed using PNG optimization and JPEG quality compression (90%) to reduce file size while keeping good quality.
Languages
The dataset is monolingual:
English (en): Texture titles and tags… See the full description on the dataset page: https://huggingface.co/datasets/nyuuzyou/cc0-textures.NFT-70M_text
Dataset Card for "NFT-70M_text"
Dataset summary
The NFT-70M_text dataset is a companion for our released NFT-70M_transactions dataset,
which is the largest and most up-to-date collection of Non-Fungible Tokens (NFT) transactions between 2021 and 2023 sourced from OpenSea.
As we also reported in the "Data anonymization" section of the dataset card of NFT-70M_transactions,
the textual contents associated with the NFT data were replaced by identifiers to numerical… See the full description on the dataset page: https://huggingface.co/datasets/MLNTeam-Unical/NFT-70M_text.text-2-image-dpo-human-preferences-full
Text-2-Image DPO Human Preferences (Full)
The complete human preference dataset for text-to-image generation. 416,360 pairwise judgments from ~20,000 annotators comparing AI-generated images across two evaluation dimensions: prompt alignment and overall preference.
This is the full, unfiltered version with uniform vote weights. For quality-filtered subsets with calibrated annotator weighting, see:
datapointai/text-2-image-dpo-human-preferences (5,000 pairs, trust-weighted)… See the full description on the dataset page: https://huggingface.co/datasets/datapointai/text-2-image-dpo-human-preferences-full.texture-shape-cue-conflictThis dataset contains the stimuli for ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness by Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A. Wichmann, and Wieland Brendel.
The stimuli allow testing of the texture/shape bias in an observer model (human or artificial) by containing two conflicting cues per image (shape and texture). The images generated using iterative style transfer (Gatys et al., 2016) between… See the full description on the dataset page: https://huggingface.co/datasets/rgeirhos/texture-shape-cue-conflict.Synth-Text-Eng-512x128
Synthetic Text Images (English)
A synthetic dataset of rendered text images with rich per-sample
annotations: the text itself, its rendering attributes, background
description, applied post-processing, and a natural-language caption.
Each image is generated by compositing English text over a procedurally
generated background with random font, color, position, rotation, blur,
brightness and noise. All samples are accompanied by a structured
metadata.csv and a ready-to-use… See the full description on the dataset page: https://huggingface.co/datasets/Nininkkka/Synth-Text-Eng-512x128.textureninja
Dataset Card for Texture Ninja
Dataset Summary
This dataset contains 4,540 texture images from texture.ninja. It includes high-resolution textures of brick, concrete, rock, wood, metal, paint, plaster, ground materials, and other surfaces. The original images were downloaded, processed, and compressed using PNG optimization and JPEG quality compression (90%) to reduce file size while maintaining good quality.
Languages
The dataset is monolingual:
English (en):… See the full description on the dataset page: https://huggingface.co/datasets/nyuuzyou/textureninja.text-to-image-2M
text-to-image-2M: A High-Quality, Diverse Text-to-Image Training Dataset
Overview
text-to-image-2M is a curated text-image pair dataset designed for fine-tuning text-to-image models. The dataset consists of approximately 2 million samples, carefully selected and enhanced to meet the high demands of text-to-image model training. The motivation behind creating this dataset stems from the observation that datasets with over 1 million samples tend to produce better… See the full description on the dataset page: https://huggingface.co/datasets/rivisia/text-to-image-2M.texturecan
Dataset Card for TextureCan Textures
Dataset Summary
This dataset contains 4,037 texture images from texturecan.com. It includes textures of various materials such as brick, paper, fabric, metal, wood, stone, and other surfaces. The original archives were downloaded, unpacked, and images were compressed using PNG optimization and JPEG quality compression (90%) to reduce file size while maintaining good quality.
Languages
The dataset is monolingual:
English… See the full description on the dataset page: https://huggingface.co/datasets/nyuuzyou/texturecan.text-2-image-dpo-human-preferences
Text-2-Image DPO Human Preferences
A large-scale, quality-controlled human preference dataset for text-to-image generation. 80,000 trust-weighted pairwise judgments from calibrated annotators comparing AI-generated images across two evaluation dimensions: prompt alignment and overall preference.
Built on the Datapoint annotation platform — purpose-built infrastructure for collecting high-quality human preference data at scale.
Overview
Metric
Value
Total… See the full description on the dataset page: https://huggingface.co/datasets/datapointai/text-2-image-dpo-human-preferences.text-to-image-2M
text-to-image-2M: A High-Quality, Diverse Text-to-Image Training Dataset
Overview
text-to-image-2M is a curated text-image pair dataset designed for fine-tuning text-to-image models. The dataset consists of approximately 2 million samples, carefully selected and enhanced to meet the high demands of text-to-image model training. The motivation behind creating this dataset stems from the observation that datasets with over 1 million samples tend to produce better… See the full description on the dataset page: https://huggingface.co/datasets/Junaid7188/text-to-image-2M.text-2-image-dpo-human-preferences-small
Text-2-Image DPO Human Preferences (Small)
A quality-controlled human preference dataset for text-to-image generation. 40,000 trust-weighted pairwise judgments from calibrated annotators comparing AI-generated images across two evaluation dimensions: prompt alignment and overall preference.
This is the highest-annotator-quality subset. For the full 5,000-pair dataset, see datapointai/text-2-image-dpo-human-preferences.
Built on the Datapoint annotation platform — purpose-built… See the full description on the dataset page: https://huggingface.co/datasets/datapointai/text-2-image-dpo-human-preferences-small.