A (PyTorch) imbalanced dataset sampler for oversampling low frequent classes and undersampling high frequent ones.

Last update: Jan 08, 2023

Overview

Imbalanced Dataset Sampler

Introduction

In many machine learning applications, we often come across datasets where some types of data may be seen more than other types. Take identification of rare diseases for example, there are probably more normal samples than disease ones. In these cases, we need to make sure that the trained model is not biased towards the class that has more data. As an example, consider a dataset where there are 5 disease images and 20 normal images. If the model predicts all images to be normal, its accuracy is 80%, and F1-score of such a model is 0.88. Therefore, the model has high tendency to be biased toward the ‘normal’ class.

To solve this problem, a widely adopted technique is called resampling. It consists of removing samples from the majority class (under-sampling) and / or adding more examples from the minority class (over-sampling). Despite the advantage of balancing classes, these techniques also have their weaknesses (there is no free lunch). The simplest implementation of over-sampling is to duplicate random records from the minority class, which can cause overfitting. In under-sampling, the simplest technique involves removing random records from the majority class, which can cause loss of information.

resampling

In this repo, we implement an easy-to-use PyTorch sampler ImbalancedDatasetSampler that is able to

rebalance the class distributions when sampling from the imbalanced dataset
estimate the sampling weights automatically
avoid creating a new balanced dataset
mitigate overfitting when it is used in conjunction with data augmentation techniques

Usage

For a simple start install the package via one of following ways:

pip install https://github.com/ufoym/imbalanced-dataset-sampler/archive/master.zip

Simply pass an ImbalancedDatasetSampler for the parameter sampler when creating a DataLoader. For example:

from torchsampler import ImbalancedDatasetSampler

train_loader = torch.utils.data.DataLoader(
    train_dataset,
    sampler=ImbalancedDatasetSampler(train_dataset),
    batch_size=args.batch_size,
    **kwargs
)

Then in each epoch, the loader will sample the entire dataset and weigh your samples inversely to your class appearing probability.

Example: Imbalanced MNIST Dataset

Distribution of classes in the imbalanced dataset:

With Imbalanced Dataset Sampler:

(left: test acc in each epoch; right: confusion matrix)

Without Imbalanced Dataset Sampler:

(left: test acc in each epoch; right: confusion matrix)

Note that there are significant improvements for minor classes such as 2 6 9, while the accuracy of the other classes is preserved.

Contributing

We appreciate all contributions. If you are planning to contribute back bug-fixes, please do so without any further discussion. If you plan to contribute new features, utility functions or extensions, please first open an issue and discuss the feature with us.

Licensing

MIT licensed.

Comments

NotImplemented Error while running ImbalancedDatasetSampler

I followed the steps exactly according to the readme file. Yet I am getting a notimplemented error. There's no explanation for the error as well.

Here's my code: `from torchvision import transforms from torchsampler import ImbalancedDatasetSampler

batch_size = 128 val_split = 0.2 shuffle_dataset=True random_seed=42

dataset_size = len(melanoma_dataset) indices = list(range(dataset_size)) split = int(np.floor(val_split * dataset_size)) if shuffle_dataset : np.random.seed(random_seed) np.random.shuffle(indices) train_indices, test_indices = indices[split:], indices[:split]

train_loader = torch.utils.data.DataLoader(melanoma_dataset,batch_size=batch_size,sampler=ImbalancedDatasetSampler(melanoma_dataset)) test_loader = torch.utils.data.DataLoader(melanoma_dataset,batch_size=batch_size,sampler=test_sampler)`

opened by aryamansriram 8

'MyDataset' object has no attribute 'get_labels'

When I try to use my own Dataset class, I get the error 'MyDataset' object has no attribute 'get_labels' and cannot proceed.

The content of the Dataloader is as follows, and there is nothing strange about it. It processes the image data and label data in .npz format.

class MyDataset(data.Dataset):
    def __init__(self, images, labels, transform=None):
        self.images = images
        self.labels = labels
        self.transform = transform

    def __len__(self):
        return len(self.images)

    def __getitem__(self, index):
        image = self.images[index]
        label = self.labels[index]

        if self.transform is not None:
            image = self.transform(image=image)["image"]

        return image, label

train_dataset = MyDataset(train_imgs, train_labels, transform=transform)
train_dataloader = torch.utils.data.DataLoader(train_dataset,
                                               sampler=ImbalancedDatasetSampler(train_dataset),
                                               batch_size= batch_size,
                                               shuffle=True,
                                               num_workers=2)

Is there something wrong with the code? I don't think it's a typo.

How can I fix it so that it works correctly?

opened by kuri54 5

Reduce time for handling large imbalanced dataset

In the current code, it has lots of costs for calculating weights and statistics of the number of each label when calling init(). It might be a problem when incoming a large dataset, so I changed this part using pandas. I checked it reduces time within 1 second when I use over 5 million data. (In previous code, it takes 20 minutes to use a sampler)

opened by hwany-j 3
pip install error

Follows your method to install the package: git clone https://github.com/ufoym/imbalanced-dataset-sampler.git cd imbalanced-dataset-sampler python setup.py install pip install .

But when I run "pip install .", I met the error as follows: FileNotFoundError: [Errno 2] No such file or directory: '/home/miniconda3/envs/pytorch/lib/python3.7/site-packages/torchsampler-0.1-py3.7.egg'

How can I resolve it?

opened by wly-ai-bj 3
Publish package to PyPi and GitHub Releases
Here at https://github.com/neuropoly/ we think your project is very useful and would like to build on it! Unfortunately if we write a setup.cfg for ourselves that depends on "torchsampler", it fails to install, because https://pypi.org/project/torchsampler/ is a 404:

This Workflow script makes publishing the conventional way easy and reliable, and will mean that projects like ours can build on your work!

To use it,

Create an account at https://pypi.org/

Go to https://pypi.org/manage/account/token/ and make a token

Go to then go to https://github.com/kousu/imbalanced-dataset-sampler/settings/secrets/actions and paste the token in there, with name "PYPI_TOKEN"

Then every time you are ready to publish,

Go to https://github.com/ufoym/imbalanced-dataset-sampler/releases/new,

Fill in a new tag like "1.0.0"

Click Publish.

It will run the Action and then in a minute or so will show up on https://github.com/kousu/imbalanced-dataset-sampler/releases/ and https://pypi.org/project/torchsampler/#history

For your first few runs, while you get used to this script, I recommend using "rc" suffixes to post versions without committing to them. For example, I've used this script myself to produce

the ones with the yellow "pre-release" tags get ignored by pip by default, unless a user opts into them with pip install --pre.

Publishing to PyPI will avoid issues like

https://github.com/ufoym/imbalanced-dataset-sampler/issues/20

https://github.com/ufoym/imbalanced-dataset-sampler/issues/15

https://github.com/ufoym/imbalanced-dataset-sampler/issues/12

and again, let us build on your work. Without publishing to pypi, it means our install instructions need to include yours:

pip install https://github.com/ufoym/imbalanced-dataset-sampler/archive/master.zip neuropoly-seg-model

which is pretty awkward for us.

Thanks a lot for your work, it's really helping out our research prototyping.
opened by kousu 2
latest commit cannot get correct labels from ImageFolder dataset
the ImageFolder.imgs return a list of 2-element tuple, such as:

[ ('classA/img01.jpg', 0), ('classB/img01.jpg', 1), ... ('classN/img01.jpg', N-1) ]

if we use dataset.imgs[:][1], it cannot not get correct labels of all samples.

at this line: https://github.com/ufoym/imbalanced-dataset-sampler/blob/756b0b61ca48a026e9a5c216296a520a10faf7df/torchsampler/imbalanced.py#L46
opened by TriLoo 2
reduce the time when calling sampler

In the current code, it has lots of costs for calculating weights and statistics of the number of each label when calling init(). It might be a problem when incoming a large dataset, so I changed this part using pandas. I checked it reduces time within 1 second when I use over 5 million data. (In previous code, it takes 20 minutes to use a sampler)

opened by hwany-j 2
Use setuptools_scm
Follows up https://github.com/ufoym/imbalanced-dataset-sampler/pull/47

on pypi it's 0.1.1

on github it's 0.1.0

but the attached wheel and sdist are 0.1.1

setuptools-scm makes git the single-source of truth for the version of the package, and works well with the build script in #47 (which is triggered by tagging a new version in git).

The other thing setuptools-scm does is make git the single source of truth for your MANIFEST, so this drops check-manifest and also most of the contents of your MANIFEST.in -- except the prune lines, those are still doing something for the moment.
opened by kousu 1
Too much time cost

It slower than before too many times when I use this sampler

(self.indices[i] for i in torch.multinomial(self.weights, self.num_samples, replacement=True)) it seems that this expression cost too much time!

Any one have any solution?

opened by nothingeasy 1
AttributeError: 'ConcatDataset' object has no attribute 'img_norm_cfg'

when i run test.py, there is an error: File "tools/test.py", line 211, in
main()
File "tools/test.py", line 181, in main
outputs = single_gpu_test(model, data_loader, args.show, args.log_dir)
File "tools/test.py", line 39, in single_gpu_test
model.module.show_result(data, result, dataset.img_norm_cfg, dataset='DOTA1_5') AttributeError: 'ConcatDataset' object has no attribute 'img_norm_cfg'

How can I solve this problem?

opened by ZSX2018 0
I think solve issue 32

https://github.com/ufoym/imbalanced-dataset-sampler/issues/32

this issue, return one element when ImageFolder, so, i changed return part

I hope this helps.

opened by LeeTaeHoon97 0
ValueError: Cannot set a frame with no defined index and a value that cannot be converted to a Series
Hi, I am using BERT for multi label classification. The dataset is imbalance and I use ImbalancedDatasetSampler as the sampler.

The train data has been tokenized, has id, mask and label:

(tensor([ 101, 112, 872, 4761, 6887, 1914, 840, 1914, 7353, 6818, 3300, 784, 720, 1408, 136, 1506, 1506, 3300, 4788, 2357, 5456, 119, 119, 119, 4696, 4638, 741, 677, 1091, 4638, 872, 1420, 1521, 119, 119, 119, 872, 2157, 6929, 1779, 4788, 2357, 3221, 686, 4518, 677, 3297, 1920, 4638, 4788, 2357, 117, 1506, 1506, 117, 7745, 872, 4638, 1568, 2124, 3221, 6432, 2225, 1217, 2861, 4478, 4105, 2357, 3221, 686, 4518, 677, 3297, 1920, 4638, 4105, 2357, 1568, 119, 119, 119, 1506, 1506, 1506, 112, 112, 4268, 4268, 117, 1961, 4638, 1928, 1355, 5456, 106, 2769, 812, 1920, 2812, 7370, 3488, 2094, 6963, 6206, 5436, 677, 3341, 2769, 4692, 1168, 3312, 1928, 5361, 7027, 3300, 1928, 1355, 119, 119, 119, 671, 2137, 3221, 8584, 809, 1184, 1931, 1168, 4638, 117, 872, 6432, 3221, 679, 3221, 136, 138, 4495, 4567, 140, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]), tensor([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]), tensor(0))

When using

from torch.utils.data import DataLoader, RandomSampler, SequentialSampler

they are fine:

batch_size=3 dataloader_train_o = DataLoader( dataset_train, sampler=RandomSampler(dataset_train), batch_size=batch_size, # **kwargs )

However, replace the sampler to ImbalancedDatasetSampler

batch_size=3 dataloader_train_o = DataLoader( dataset_train, sampler=ImbalancedDatasetSampler(dataset_train), batch_size=batch_size, # **kwargs )

The error print below

ValueError Traceback (most recent call last) File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\frame.py:3892, in DataFrame._ensure_valid_index(self, value) 3891 try: -> 3892 value = Series(value) 3893 except (ValueError, NotImplementedError, TypeError) as err:

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\series.py:451, in Series.init(self, data, index, dtype, name, copy, fastpath) 450 else: --> 451 data = sanitize_array(data, index, dtype, copy) 453 manager = get_option("mode.data_manager")

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\construction.py:601, in sanitize_array(data, index, dtype, copy, raise_cast_failure, allow_2d) 599 subarr = maybe_infer_to_datetimelike(subarr) --> 601 subarr = _sanitize_ndim(subarr, data, dtype, index, allow_2d=allow_2d) 603 if isinstance(subarr, np.ndarray): 604 # at this point we should have dtype be None or subarr.dtype == dtype

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\construction.py:652, in _sanitize_ndim(result, data, dtype, index, allow_2d) 651 return result --> 652 raise ValueError("Data must be 1-dimensional") 653 if is_object_dtype(dtype) and isinstance(dtype, ExtensionDtype): 654 # i.e. PandasDtype("O")

ValueError: Data must be 1-dimensional

The above exception was the direct cause of the following exception:

ValueError Traceback (most recent call last) Input In [49], in <cell line: 5>() 2 from torchsampler import ImbalancedDatasetSampler 4 batch_size=3 5 dataloader_train_o = DataLoader( 6 dataset_train, ----> 7 sampler=ImbalancedDatasetSampler(dataset_train), 8 batch_size=batch_size, 9 # **kwargs 10 ) 12 dataloader_validation_o = DataLoader( 13 dataset_val, 14 sampler=SequentialSampler(dataset_val), 15 batch_size=batch_size, 16 # **kwargs 17 )

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\torchsampler\imbalanced.py:37, in ImbalancedDatasetSampler.init(self, dataset, labels, indices, num_samples, callback_get_label) 35 # distribution of classes in the dataset 36 df = pd.DataFrame() ---> 37 df["label"] = self._get_labels(dataset) if labels is None else labels 38 df.index = self.indices 39 df = df.sort_index()

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\frame.py:3655, in DataFrame.setitem(self, key, value) 3652 self._setitem_array([key], value) 3653 else: 3654 # set column -> 3655 self._set_item(key, value)

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\frame.py:3832, in DataFrame._set_item(self, key, value) 3822 def _set_item(self, key, value) -> None: 3823 """ 3824 Add series to DataFrame in specified column. 3825 (...) 3830 ensure homogeneity. 3831 """ -> 3832 value = self._sanitize_column(value) 3834 if ( 3835 key in self.columns 3836 and value.ndim == 1 3837 and not is_extension_array_dtype(value) 3838 ): 3839 # broadcast across multiple columns if necessary 3840 if not self.columns.is_unique or isinstance(self.columns, MultiIndex):

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\frame.py:4528, in DataFrame._sanitize_column(self, value) 4515 def _sanitize_column(self, value) -> ArrayLike: 4516 """ 4517 Ensures new columns (which go into the BlockManager as new blocks) are 4518 always copied and converted into an array. (...) 4526 numpy.ndarray or ExtensionArray 4527 """ -> 4528 self._ensure_valid_index(value) 4530 # We should never get here with DataFrame value 4531 if isinstance(value, Series):

File D:\ProgramData\Anaconda3\envs\pytorch\lib\site-packages\pandas\core\frame.py:3894, in DataFrame._ensure_valid_index(self, value) 3892 value = Series(value) 3893 except (ValueError, NotImplementedError, TypeError) as err: -> 3894 raise ValueError( 3895 "Cannot set a frame with no defined index " 3896 "and a value that cannot be converted to a Series" 3897 ) from err 3899 # GH31368 preserve name of index 3900 index_copy = value.index.copy()

ValueError: Cannot set a frame with no defined index and a value that cannot be converted to a Series
opened by zhangjizby 1

Implementation for Pytorch-geometric dataset

I have added a few lines that allow to work with pytorch-geometric dataset. Since Pytorch-geometric data is saved as a list before being loaded by a Pytorch-geometric Dataloader, the modification is pretty simple. Hope this could be helpful to someone.

Best,

Anna

`from typing import Callable

import pandas as pd import torch import torch.utils.data import torchvision

class ImbalancedDatasetSampler(torch.utils.data.sampler.Sampler): """Samples elements randomly from a given list of indices for imbalanced dataset

Arguments:
    indices: a list of indices
    num_samples: number of samples to draw
    callback_get_label: a callback-like function which takes two arguments - dataset and index
"""

def __init__(self, dataset, indices: list = None, num_samples: int = None, callback_get_label: Callable = None):
    # if indices is not provided, all elements in the dataset will be considered
    self.indices = list(range(len(dataset))) if indices is None else indices

    # define custom callback
    self.callback_get_label = callback_get_label

    # if num_samples is not provided, draw `len(indices)` samples in each iteration
    self.num_samples = len(self.indices) if num_samples is None else num_samples

    # distribution of classes in the dataset
    df = pd.DataFrame()
    df["label"] = self._get_labels(dataset)
    df.index = self.indices
    df = df.sort_index()

    label_to_count = df["label"].value_counts()

    weights = 1.0 / label_to_count[df["label"]]

    self.weights = torch.DoubleTensor(weights.to_list())

def _get_labels(self, dataset):
    if self.callback_get_label:
        return self.callback_get_label(dataset)
    elif isinstance(dataset, torchvision.datasets.MNIST):
        return dataset.train_labels.tolist()
    elif isinstance(dataset, torchvision.datasets.ImageFolder):
        return [x[1] for x in dataset.imgs]
    elif isinstance(dataset, torchvision.datasets.DatasetFolder):
        return dataset.samples[:][1]
    elif isinstance(dataset, torch.utils.data.Subset):
        return dataset.dataset.imgs[:][1]
    elif isinstance(dataset, torch.utils.data.Dataset):
        return dataset.get_labels()
    elif isinstance(dataset, list):
        return [dataset[i].y.item() for i in range(len(dataset))]  #here the modification
    else:
        raise NotImplementedError

def __iter__(self):
    return (self.indices[i] for i in torch.multinomial(self.weights, self.num_samples, replacement=True))

def __len__(self):
    return self.num_samples

opened by avarbella 0

callback_get_label no longer works as described

callback_get_label: a callback-like function which takes two arguments - dataset and index

This no longer seems to be the case?

Please update how the new use-case looks like, because above commit breaks a lot of my previous code.

@hwany-j

opened by zimonitrome 3

Releases(v0.1.2)

v0.1.2(May 23, 2022)

Source code(tar.gz)
Source code(zip)
torchsampler-0.1.2-py3-none-any.whl(5.47 KB)
torchsampler-0.1.2.tar.gz(6.44 KB)
v0.1.0(May 18, 2022)

Source code(tar.gz)
Source code(zip)
torchsampler-0.1.1-py3-none-any.whl(5.29 KB)
torchsampler-0.1.1.tar.gz(5.63 KB)

Owner

Ming

[email protected] ->

GitHub Repository

Source code for NAACL 2021 paper "TR-BERT: Dynamic Token Reduction for Accelerating BERT Inference"

TR-BERT Source code and dataset for "TR-BERT: Dynamic Token Reduction for Accelerating BERT Inference". The code is based on huggaface's transformers.

37 Oct 30, 2022

Code and data form the paper BERT Got a Date: Introducing Transformers to Temporal Tagging

BERT Got a Date: Introducing Transformers to Temporal Tagging Satya Almasian*, Dennis Aumiller*, and Michael Gertz Heidelberg University Contact us vi

54 Dec 04, 2022

BisQue is a web-based platform designed to provide researchers with organizational and quantitative analysis tools for 5D image data. Users can extend BisQue by implementing containerized ML workflows.

Overview BisQue is a web-based platform specifically designed to provide researchers with organizational and quantitative analysis tools for up to 5D

26 Nov 29, 2022

Automatically download the cwru data set, and then divide it into training data set and test data set

Automatically download the cwru data set, and then divide it into training data set and test data set.自动下载cwru数据集，然后分训练数据集和测试数据集

6 Jun 27, 2022

Pytorch implementation AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative Adversarial Networks

AttnGAN Pytorch implementation for reproducing AttnGAN results in the paper AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative

1.2k Dec 26, 2022

[AAAI22] Reliable Propagation-Correction Modulation for Video Object Segmentation

Reliable Propagation-Correction Modulation for Video Object Segmentation (AAAI22) Preview version paper of this work is available at: https://arxiv.or

70 Dec 04, 2022

用强化学习DQN算法，训练AI模型来玩合成大西瓜游戏，提供Keras版本和PARL（paddle）版本

用强化学习玩合成大西瓜代码地址：https://github.com/Sharpiless/play-daxigua-using-Reinforcement-Learning 用强化学习DQN算法，训练AI模型来玩合成大西瓜游戏，提供Keras版本、PARL（paddle）版本和pytorch版本

72 Dec 17, 2022

Face Mask Detector by live camera using tensorflow-keras, openCV and Python

Face Mask Detector 😷 by Live Camera Detecting masked or unmasked faces by live camera with percentange of mask occupation About Project: This an Arti

2 Apr 04, 2022

A proof of concept ai-powered Recaptcha v2 solver

Recaptcha Fullauto I've decided to open source my old Recaptcha v2 solver. My latest version will be opened sourced this summer. I am hoping this proj

60 Dec 20, 2022

This tool uses Deep Learning to help you draw and write with your hand and webcam.

This tool uses Deep Learning to help you draw and write with your hand and webcam. A Deep Learning model is used to try to predict whether you want to have 'pencil up' or 'pencil down'.

169 Dec 10, 2022

G-NIA model from "Single Node Injection Attack against Graph Neural Networks" (CIKM 2021)

Single Node Injection Attack against Graph Neural Networks This repository is our Pytorch implementation of our paper: Single Node Injection Attack ag

18 Nov 21, 2022

Flaxformer: transformer architectures in JAX/Flax

Flaxformer is a transformer library for primarily NLP and multimodal research at Google.

116 Jan 05, 2023

Lightweight plotting to the terminal. 4x resolution via Unicode.

Uniplot Lightweight plotting to the terminal. 4x resolution via Unicode. When working with production data science code it can be handy to have plotti

203 Dec 29, 2022

FuseDream: Training-Free Text-to-Image Generationwith Improved CLIP+GAN Space OptimizationFuseDream: Training-Free Text-to-Image Generationwith Improved CLIP+GAN Space Optimization

FuseDream This repo contains code for our paper (paper link): FuseDream: Training-Free Text-to-Image Generation with Improved CLIP+GAN Space Optimizat

191 Dec 31, 2022