datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
BLIP3o-Pretrain-Long-Caption
BLIP3o Pretrain Long-Caption Dataset
This collection contains 27 million images, each paired with a long (~120 token) caption generated by Qwen/Qwen2.5-VL-7B-Instruct.
Download
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="BLIP3o/BLIP3o-Pretrain-Long-Caption",
repo_type="dataset"
)
Load Dataset without Extracting
You don’t need to unpack the .tar archives, use WebDataset support in 🤗datasets instead:
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/BLIP3o/BLIP3o-Pretrain-Long-Caption.BLIP3o-Pretrain-Short-Caption
BLIP3o Pretrain Short-Caption Dataset
This collection contains 5 million images, each paired with a short (~20 token) caption generated by Qwen/Qwen2.5-VL-7B-Instruct.
Download
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="BLIP3o/BLIP3o-Pretrain-Short-Caption",
repo_type="dataset"
)
Load Dataset without Extracting
You don’t need to unpack the .tar archives, use WebDataset support in 🤗datasets instead:
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/BLIP3o/BLIP3o-Pretrain-Short-Caption.blip3-ocr-200m
BLIP3-OCR-200M Dataset
Overview
The BLIP3-OCR-200M dataset is designed to address the limitations of current Vision-Language Models (VLMs) in processing and interpreting text-rich images, such as documents and charts. Traditional image-text datasets often struggle to capture nuanced textual information, which is crucial for tasks requiring complex text comprehension and reasoning.
Key Features
OCR Integration: The dataset incorporates Optical Character… See the full description on the dataset page: https://huggingface.co/datasets/Salesforce/blip3-ocr-200m.BLIP3o-Pretrain-JourneyDB
BLIP3o Pretrain JourneyDB Dataset
This collection contains 4 million JourneyDB images.
Download
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="BLIP3o/BLIP3o-Pretrain-JourneyDB",
repo_type="dataset"
)
Load Dataset without Extracting
You don’t need to unpack the .tar archives, use WebDataset support in 🤗datasets instead:
from datasets import load_dataset
import glob
data_files = glob.glob("/your/data/path/*.tar")… See the full description on the dataset page: https://huggingface.co/datasets/BLIP3o/BLIP3o-Pretrain-JourneyDB.BLIP3o-NEXT-EDIT-ENSEMBLE-DATASETSBLIP3o-60kblip3o-60kBLIP3o-60kThis is BLIP3o-60k Text-to-Image instruction tuning dataset distilled from GPT-4o, including the following categories:
JourneyDB
Human (including MSCOCO with human caption, human gestures, occupations)
Dalle3
Geneval (no overlap with test set)
Common objects
Simple text
Here we provide the code guidance to download tar file:
from huggingface_hub import snapshot_download
snapshot_download(repo_id='BLIP3o/BLIP3o-60k', repo_type=‘dataset’)
And you can use huggingface datasets to read the tar… See the full description on the dataset page: https://huggingface.co/datasets/BLIP3o/BLIP3o-60k.blip3o-256BLIP3o-JourneyDBblip3o_aux_full_8640
BLIP3o auxiliary full protocol subset (8,640 rows)
This is the 10x full auxiliary view for the current UMM-SFT experiments.
The corresponding tiny view contains 864 rows.
Rows published: 8,640
Source pool: 51,758 verified BLIP3o auxiliary rows
Selection seed: 178430
Selection details: formal_selection_receipt.json
Features: image, prompt, source
“Full” means the 10x protocol subset, not the complete upstream corpus. The
source metadata did not declare a license; other is only… See the full description on the dataset page: https://huggingface.co/datasets/xzz789/blip3o_aux_full_8640.BLIP3o-Pretrain-Long-Caption-ParquetBLIP3o-Shortblip3o-caption-mini-arrow
blip3o-caption-mini-arrow
blip3o-caption-mini-arrow is a high-quality, curated image-caption dataset derived and optimized from the original BLIP3o/BLIP3o-Pretrain-Long-Caption. This dataset is specifically filtered and processed for tasks involving long-form image captioning and vision-language understanding.
Overview
Total Samples: 91,600
Modality: Image ↔ Text
Format: Arrow (auto-converted to Parquet)
License: Apache 2.0
Language: English
Size: ~4.5 GB… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/blip3o-caption-mini-arrow.BLIP3o-Visual-Reasoningblip3o_aux_formal_864
BLIP3o auxiliary subset (864 rows)
This repository contains the deterministic Generation-auxiliary view used by
the current four-family umm-sft paired_ug_formal recipes (BAGEL, BLIP3o,
Emu3.5, and Janus-Pro).
Rows published: 864
Original verified source pool on rz6000: 51,758 rows
Selection seed: 178430
Selection details: formal_selection_receipt.json
Features: image, prompt, source
This is a protocol-specific subset, not the full BLIP3o-60k release. The
source metadata did… See the full description on the dataset page: https://huggingface.co/datasets/xzz789/blip3o_aux_formal_864.BLIP3o-LongBLIP3o-Pretrain-Long-Caption-Text-Splitsblip3o-long-caption-50k
BLIP3o Long Caption Dataset (50K Subset)
This dataset is a subset of the BLIP3o/BLIP3o-Pretrain-Long-Caption dataset, containing the first 50,000 samples.
Statistics
Total samples: 50,000
Average caption length: 624.4 characters
Min caption length: 125 characters
Max caption length: 983 characters
Dataset Structure
Data Fields
id: Unique identifier for each sample
caption: Long-form image caption
caption_length: Length of the caption in characters… See the full description on the dataset page: https://huggingface.co/datasets/russwang/blip3o-long-caption-50k.BLIP3o-60k_wan_i2v_14b_480p_512_512_81_rewriteBLIP3o-60k_1024h_1776wBLIP3o-60kBLIP3o-60k-top100BLIP3o-60k-top100-datablip3o-pretrain-short-recaptionedBLIP3o-60k-1Bagel-new-BLIP3o-5k