metadata
dataset_info:
features:
- name: dataset
dtype: string
- name: image_id
dtype: int64
- name: image
dtype: image
- name: alt_text
dtype: string
- name: license
dtype: string
- name: source
dtype: string
- name: inclusive_alt_text
dtype: string
- name: need_training
dtype:
class_label:
names:
'0': 'no'
'1': 'yes'
- name: verified
dtype:
class_label:
names:
'0': 'no'
'1': 'yes'
- name: rejection_reasons
sequence: string
- name: added_by
dtype: string
- name: verified_by
dtype: string
- name: modified_date
dtype: timestamp[ns]
- name: nsfw
dtype:
class_label:
names:
'0': 'no'
'1': 'yes'
- name: golden
dtype:
class_label:
names:
'0': 'no'
'1': 'yes'
- name: gpt_alt_text
dtype: string
splits:
- name: train
num_bytes: 1184791040.95
num_examples: 2970
download_size: 1163757100
dataset_size: 1184791040.95
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
This dataset contains images and alt text from various sources.
It is used to control the quality of https://huggingface.co/Mozilla/distilvit using the https://github.com/mozilla/checkvite application
This application let users try out the model on the images and classify them. The dataset is then updated.
When an image is marked as need_training it will be use to fine-tune the model to fix some of its inaccuracies