datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Imagenet21K_RecaptionThis dataset is the entire 21K ImageNet dataset with about 13 million examples and about 19 thousand classes as strings
(for some reason it only had ~19K classes instead of 21K).
If you want an even larger set of images, I have a recaptioned CC12M and ImageNet dataset: https://huggingface.co/datasets/gmongaras/CC12M_and_Imagenet21K_Recap
The images are in PNG format. They can be decoded like in the following example
import io
from PIL import Image
Image.open(io.BytesIO(row["image"]))
where… See the full description on the dataset page: https://huggingface.co/datasets/gmongaras/Imagenet21K_Recaption.Stable_Diffusion_3_RecaptionThis dataset is the one specified in the stable diffusion 3 paper which is composed of the ImageNet dataset and the CC12M dataset.
I used the ImageNet 2012 train/val data and captioned it as specified in the paper: "a photo of a 〈class name〉" (note all ids are 999,999,999)
CC12M is a dataset with 12 million images created in 2021. Unfortunately the downloader provided by Google has many broken links and the download takes forever.
However, some people in the community publicized the dataset.… See the full description on the dataset page: https://huggingface.co/datasets/gmongaras/Stable_Diffusion_3_Recaption.wikiart_recaptionWikiArt Dataset captioned using vikhyatk/moondream2 model with prompt : Generate a short, simple and only visually descriptive caption for this image.
COCONut-PanCap-Recaptioned
COCONut-PanCap Re-captioned (train2017)
Regenerated captions for the train2017 split of COCONut-PanCap
(paper,
dataset).
The released captions contain phrases bound to the wrong segment, references to objects that
are not in the image, redundant restatements, and untagged summary sentences that carry no
grounding. We regenerated every caption and repaired the remaining errors, keeping the file
layout of the original release.
Only the caption text changed. Mask ids… See the full description on the dataset page: https://huggingface.co/datasets/Panorama-grounding/COCONut-PanCap-Recaptioned.wit_1m_recaptioned
Back into Plato’s Cave: Examining Cross-modal Representational Convergence at Scale
Image–text dataset derived from Wikipedia-based Image Text (WIT) with original and Gemini-generated captions, introduced in Back into Plato’s Cave: Examining Cross-modal Representational Convergence at Scale.
Configs
wit_1024
A fixed set of 1,024 query samples used for alignment evaluation.
from datasets import load_dataset
ds = load_dataset("askoepke/wit_1m_recaptioned"… See the full description on the dataset page: https://huggingface.co/datasets/askoepke/wit_1m_recaptioned.megalith-10m-5.5k-claude-opus-5-recaptioned
Megalith-10M 5.5K — Claude Opus 5 Recaptioned
This is a 5,511-image derivative subset of madebyollin/megalith-10m, selected through the megalith10m portion of zlab-princeton/i1-captions. The bytes were retrieved from the drawthingsai/megalith-10m image archive. It is not the complete Megalith-10M collection.
Every image has one newly generated, detailed English caption. The recaptioning was performed with Claude Opus 5 via Claude Code on August 2, 2026. The image was the primary… See the full description on the dataset page: https://huggingface.co/datasets/sirus/megalith-10m-5.5k-claude-opus-5-recaptioned.inaturalist-2024-2.8k-claude-opus-5-recaptioned
iNaturalist 2024 2.8K — Claude Opus 5 Recaptioned
This is a 2,824-image derivative subset of iNaturalist 2024 (iNat24), distributed through the INQUIRE project, selected through the inaturalist portion of zlab-princeton/i1-captions. It is not the complete 4.8-million-image iNat24 training set.
Every image has one newly generated, detailed English caption. The recaptioning was performed with Claude Opus 5 via Claude Code on August 2, 2026. The image was the primary evidence; the… See the full description on the dataset page: https://huggingface.co/datasets/sirus/inaturalist-2024-2.8k-claude-opus-5-recaptioned.JourneyDB-recaption
JourneyDB Recaption
Recaptioned version of the JourneyDB dataset using Qwen vision-language models.
Dataset Description
JourneyDB is a large-scale dataset of AI-generated images from Midjourney. This recaptioned version provides detailed visual descriptions generated by a vision-language model, which are more accurate than the original generation prompts for describing actual image content.
Statistics
Metric
Count
Total rows
3,389,605
File size… See the full description on the dataset page: https://huggingface.co/datasets/undefined443/JourneyDB-recaption.JourneyDB-recaption
JourneyDB Recaption
Recaptioned version of the JourneyDB dataset using Qwen vision-language models.
Dataset Description
JourneyDB is a large-scale dataset of AI-generated images from Midjourney. This recaptioned version provides detailed visual descriptions generated by a vision-language model, which are more accurate than the original generation prompts for describing actual image content.
Statistics
Metric
Count
Total rows
3,389,605… See the full description on the dataset page: https://huggingface.co/datasets/toilaluan/JourneyDB-recaption.datikz-v3-recaptioned
DaTikZ-v3 Recaptioned
A 10,000-sample subset of DaTikZ-v3 with captions regenerated using Gemini Flash via vision-language captioning.
Captioning
Original captions were replaced by passing each rendered diagram image to Gemini Flash with the prompt: describe the diagram concisely, focusing on geometric shapes, mathematical concepts, key visual elements, and purpose (~1–3 sentences starting with "A diagram ..."). Only samples with TikZ code shorter than 1000 characters… See the full description on the dataset page: https://huggingface.co/datasets/catslashbin/datikz-v3-recaptioned.cc12m-wds-recaption
CC12M with Enhanced Captions
This dataset contains 1.3 million image-text pairs from the CC12M dataset with model-generated captions.
Dataset Details
Total Samples: 1,306,239
Source: pixparse/cc12m-wds
Captioning Model: Qwen/Qwen3-VL-8B-Instruct
Format: Parquet
Filtering Criteria
Samples were filtered based on the following quality metrics:
Aesthetic Score: >= 5.5 (using LAION aesthetic classifier)
Resolution: >= 512 pixels (width or height)
Aspect Ratio: <=… See the full description on the dataset page: https://huggingface.co/datasets/undefined443/cc12m-wds-recaption.laion-unsafe-downloaded-recaptionedlaion-mix-updated-recaptioned-yolo-filtered-2LAION-Art-recaption
LAION-Art Recaption
Recaptioned subset of the LAION-Art dataset using Qwen2.5-VL-7B-Instruct.
Dataset Description
This dataset contains detailed recaptions for LAION-Art images generated by a vision-language model. Only successfully recaptioned samples are included.
Total samples: 1,410,704
Columns
Column
Type
Description
url
string
Original image URL
key
string
Unique identifier for each sample
width
int32
Image width
height
int32
Image height… See the full description on the dataset page: https://huggingface.co/datasets/undefined443/LAION-Art-recaption.filtered-wit-recaptionedcc3m-recaptionedcc12m-recaptionedzangei-dit-stage-1-250k-recaptionedvisualize-recaptioned-test1
Dataset Card for "visualize-recaptioned-test1"
More Information needed
visualize-paligemma-recaptioned
Dataset Card for "visualize-paligemma-recaptioned"
More Information needed
imagenet-recaptionedzangei-dit-stage-1-250k-recaptioned-qwen-3-0.6b-embeddingsyfcc-recaptionedstock-image-recaptionmerged_aa_recaptionedllava-onevision-recaption-ko-preview