Datasets:

Languages:
English
ArXiv:
License:
ImageRewardDB / ImageRewardDB.py
wuyuchen's picture
Upload ImageRewardDB.py with huggingface_hub
84b69b4
raw
history blame
9.79 kB
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""TODO: Add a description here."""
import pandas as pd
import json
import os
import datasets
from huggingface_hub import hf_hub_url
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@misc{xu2023imagereward,
title={ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation},
author={Jiazheng Xu and Xiao Liu and Yuchen Wu and Yuxuan Tong and Qinkai Li and Ming Ding and Jie Tang and Yuxiao Dong},
year={2023},
eprint={2304.05977},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
"""
# You can copy an official description
_DESCRIPTION = """\
We systematically identify the challenges for text-to-image human preference annotation, and \
consequently design a pipeline tailored for it, establishing criteria for quantitative assessment and \
annotator training, optimizing labeling experience, and ensuring quality validation. We build this \
text-to-image comparison dataset, ImageRewardDB, for training the ImageReward model based on the pipeline.\
The ImageRewarDB covers both the rating and ranking components, collecting a dataset of 137k expert \
comparisons to date.
"""
_HOMEPAGE = "https://huggingface.co/datasets/wuyuchen/ImageRewardDB"
_VERSION = datasets.Version("1.0.0")
_LICENSE = "apache-2.0"
# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
_REPO_ID = "wuyuchen/ImageRewardDB"
_URLS = {}
_PART_IDS = {
"train": 32,
"validation": 2,
"test": 2
}
for name in list(_PART_IDS.keys()):
_URLS[name] = {}
for i in range(1, _PART_IDS[name]+1):
_URLS[name][i] = hf_hub_url(
_REPO_ID,
filename=f"images/{name}/{name}_{i}.zip",
repo_type="dataset"
)
_URLS[name]["metadata"] = hf_hub_url(
_REPO_ID,
filename=f"metadata-{name}.parquet",
repo_type="dataset"
)
class ImageRewardDBConfig(datasets.BuilderConfig):
'''BuilderConfig for ImageRewardDB'''
def __init__(self, part_ids, **kwargs):
'''BuilderConfig for ImageRewardDB
Args:
part_ids([int]): A list of part_ids.
**kwargs: keyword arguments forwarded to super
'''
super(ImageRewardDBConfig, self).__init__(version=_VERSION, **kwargs)
self.part_ids = part_ids
class ImageRewardDB(datasets.GeneratorBasedBuilder):
"""A dataset of 137k expert comparisons to date, demonstrating the text-to-image human preference."""
# This is an example of a dataset with multiple configurations.
# If you don't want/need to define several sub-sets in your dataset,
# just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
# If you need to make complex sub-parts in the datasets with configurable options
# You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
# BUILDER_CONFIG_CLASS = MyBuilderConfig
# You will be able to load one or the other configurations in the following list with
# data = datasets.load_dataset('my_dataset', 'first_domain')
# data = datasets.load_dataset('my_dataset', 'second_domain')
BUILDER_CONFIGS = []
for num_k in [1,2,4,8]:
part_ids = {
"train": 4*num_k,
"validation": 2,
"test": 2
}
BUILDER_CONFIGS.append(
ImageRewardDBConfig(name=f"{num_k}k", part_ids=part_ids, description=f"This is a {num_k}k-scale ImageRewardDB")
)
DEFAULT_CONFIG_NAME = "8k" # It's not mandatory to have a default configuration. Just use one if it make sense.
def _info(self):
features = datasets.Features(
{
"image": datasets.Image(),
"prompt_id": datasets.Value("string"),
"prompt": datasets.Value("string"),
"classification": datasets.Value("string"),
"image_amount_in_total": datasets.Value("int8"),
"rank": datasets.Value("int8"),
"overall_rating": datasets.Value("int8"),
"image_text_alignment_rating": datasets.Value("int8"),
"fidelity_rating": datasets.Value("int8")
}
)
return datasets.DatasetInfo(
# This is the description that will appear on the datasets page.
description=_DESCRIPTION,
# This defines the different columns of the dataset and their types
features=features, # Here we define them above because they are different between the two configurations
# If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
# specify them. They'll be used if as_supervised=True in builder.as_dataset.
# supervised_keys=("sentence", "label"),
# Homepage of the dataset for documentation
homepage=_HOMEPAGE,
# License for the dataset if available
license=_LICENSE,
# Citation for the dataset
citation=_CITATION,
)
def _split_generators(self, dl_manager):
# If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name
# dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
# It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
# By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
data_dirs = {name: [] for name in list(_PART_IDS.keys())}
json_paths = {name: [] for name in list(_PART_IDS.keys())}
metadata_paths = {name: [] for name in list(_PART_IDS.keys())}
for key in list(self.config.part_ids.keys()):
for i in range(1, self.config.part_ids[key]+1):
data_dir = dl_manager.download_and_extract(_URLS[key][i])
data_dirs[key].append(data_dir)
json_paths[key].append(os.path.join(data_dir, f"{key}_{i}.json"))
metadata_paths[key] = dl_manager.download(_URLS[key]["metadata"])
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "train",
"data_dirs": data_dirs["train"],
"json_paths": json_paths["train"],
"metadata_path": metadata_paths["train"]
},
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "validation",
"data_dirs": data_dirs["validation"],
"json_paths": json_paths["validation"],
"metadata_path": metadata_paths["validation"]
},
),
datasets.SplitGenerator(
name=datasets.Split.TEST,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"split": "test",
"data_dirs": data_dirs["test"],
"json_paths": json_paths["test"],
"metadata_path": metadata_paths["test"]
},
),
]
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
def _generate_examples(self, split, data_dirs, json_paths, metadata_path):
# The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
num_data_dirs = len(data_dirs)
assert num_data_dirs == len(json_paths)
#Iterate throug all extracted zip folders for images
for index, json_path in enumerate(json_paths):
json_data = json.load(open(json_path, "r", encoding="utf-8"))
for example in json_data:
image_path = os.path.join(data_dirs[index], str(example["image_path"]).split("/")[-1])
yield example["image_path"], {
"image": {
"path": image_path,
"bytes": open(image_path, "rb").read()
},
"prompt_id": example["prompt_id"],
"prompt": example["prompt"],
"classification": example["classification"],
"image_amount_in_total": example["image_amount_in_total"],
"rank": example["rank"],
"overall_rating": example["overall_rating"],
"image_text_alignment_rating": example["image_text_alignment_rating"],
"fidelity_rating": example["fidelity_rating"]
}