datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cc3m-wds
Dataset Card for Conceptual Captions (CC3M)
Dataset Summary
Conceptual Captions is a dataset consisting of ~3.3M images annotated with captions. In contrast with the curated style of other image caption annotations, Conceptual Caption images and their raw descriptions are harvested from the web, and therefore represent a wider variety of styles. More precisely, the raw descriptions are harvested from the Alt-text HTML attribute associated with web images. To arrive at the… See the full description on the dataset page: https://huggingface.co/datasets/pixparse/cc3m-wds.LLaVA-ReCap-CC3Mcc3mCC3MLLaVA-CC3M-Pretrain-595K
LLaVA Visual Instruct CC3M 595K Pretrain Dataset Card
Dataset details
Dataset type:
LLaVA Visual Instruct CC3M Pretrain 595K is a subset of CC-3M dataset, filtered with a more balanced concept coverage distribution.
Captions are also associated with BLIP synthetic caption for reference.
It is constructed for the pretraining stage for feature alignment in visual instruction tuning.
We aim to build large multimodal towards GPT-4 vision/language capability.
Dataset date:… See the full description on the dataset page: https://huggingface.co/datasets/liuhaotian/LLaVA-CC3M-Pretrain-595K.cc3m_01122022cc3mThis repo is CC3M's unofficial huggingface repo.
However, for the large picture, we process it as follow and then upload:
if pil_image.width > 1024 or pil_image.height > 1024:
pil_image = pil_image.resize((1024, 1024), Image.BICUBIC)
LLaVA-CC3M-Pretrain-595Kcc3m_tokenized
Experiments for training Auto Regressive models for text-to-image generation
This dataset is derived from conceptual captions (CC3M) which contains roughly 3.3M image and caption pairs. For images we use 1d-tokenizer by bytedance which tokenizes a 256 * 256 image into 32 tokens while still achieving SOTA fidelity ratio. For text we train a BPE based tokenizer on the image captions dataset with a vocab size set to 30K, where 4096 tokens where used to represent images, 9 to represent… See the full description on the dataset page: https://huggingface.co/datasets/darshanmakwana/cc3m_tokenized.CC3M_synthetic
Dataset Card for CC3M_synthetic
Dataset Summary
This is a data of synthetic captions for Conceptual Captions. The alt_text is the original caption in CC3M dataset. The synthetic_captions were produced by Florence-2-large.
Procedure
We captioned the images by Florence-2-large. We gave the model DETAILED_CAPTION task with a beam search size of 3. To ensure quality of the captions, we filtered out samples with too short (fewer than 50 characters) or excessively… See the full description on the dataset page: https://huggingface.co/datasets/timjeffrey10/CC3M_synthetic.cc3m-256cc3m-recap-wdsCC3M-35L
Team and Homepage
Official Website: https://aienthusiasm.vn
Hugging Face Organization: https://huggingface.co/ai-enthusiasm-community
Contact
If you encounter any issues with the dataset or have any inquiries, please feel free to reach out to us via email at: aienthusiasm.team@gmail.com
Dataset Structure
The dataset is provided in a flattened tabular format, optimized for the Hugging Face Dataset Viewer and high-speed Parquet processing.
Data Fields… See the full description on the dataset page: https://huggingface.co/datasets/ai-enthusiasm-community/CC3M-35L.LLaVA-CC3M-Pretrain-595K-Embedded
Dataset derived from liuhaotian/LLaVA-CC3M-Pretrain-595K
Dataset details
Dataset type:
LLaVA Visual Instruct CC3M Pretrain 595K is a subset of CC-3M dataset, filtered with a more balanced concept coverage distribution.
Captions are also associated with BLIP synthetic caption for reference.
It is constructed for the pretraining stage for feature alignment in visual instruction tuning.
We aim to build large multimodal towards GPT-4 vision/language capability.
cc3m-subset-100kSD_Bias_CC3MThis repository contains the training dataset for the paper Would Deep Generative Models Amplify Bias in Future Models?
This dataset contains the images generated for training OpenCLIP and for exploring how the generated training data will change the social biases in OpenCLIP.
The images are generated by inputting the CC3M's captions as the prompts towards Stable Diffusion v1.5.
llava-filtered-cc3m-595kDataset transformed to the image-caption format from https://huggingface.co/datasets/liuhaotian/LLaVA-CC3M-Pretrain-595K
cc3m-grounded-annotations
CC3M grounded annotations
Region-level grounding for Conceptual Captions 3M: bounding boxes, the noun
phrase each box grounds, and the span of the caption that phrase came from, for
3,016,640 of CC3M's 3,318,333 rows.
No images here. This is metadata only, joinable onto a CC3M copy you already
have. That is the point of it: the grounding is 354 MB, the pixels are 125 GB.
Files
file
rows
size
annotations-0000..0482.parquet
3,016,640
199 MB… See the full description on the dataset page: https://huggingface.co/datasets/freek23/cc3m-grounded-annotations.generated-image-cc3m-cc12mcc3mCC3MHindi-LLaVA-CC3M-Pretrain-595K-3cc3mcc3m-wds
Dataset Card for Conceptual Captions (CC3M)
Dataset Summary
Conceptual Captions is a dataset consisting of ~3.3M images annotated with captions. In contrast with the curated style of other image caption annotations, Conceptual Caption images and their raw descriptions are harvested from the web, and therefore represent a wider variety of styles. More precisely, the raw descriptions are harvested from the Alt-text HTML attribute associated with web images. To arrive at the… See the full description on the dataset page: https://huggingface.co/datasets/lifehacker777/cc3m-wds.cc3m_500_samplecc3m-morepiecesIt's just the unlabeled train split of datasets/conceptual_captions but split into 4 pieces
cc3m_wqLLaVA-CC3M-Pretrain-595K
LLaVA Visual Instruct CC3M 595K Pretrain Dataset Card
Dataset details
Dataset type:
LLaVA Visual Instruct CC3M Pretrain 595K is a subset of CC-3M dataset, filtered with a more balanced concept coverage distribution.
Captions are also associated with BLIP synthetic caption for reference.
It is constructed for the pretraining stage for feature alignment in visual instruction tuning.
We aim to build large multimodal towards GPT-4 vision/language capability.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/DAXIN123/LLaVA-CC3M-Pretrain-595K.rc_cc3mCC3m-embed-vicuna