triple-mu / flowvision

Datasets, Transforms and Models specific to Computer Vision

flowvision

Introduction

The flowvision package consists of popular datasets, SOTA computer vision models, layers, utilities, schedulers, advanced data augmentations and common image transformations based on OneFlow.

Installation

First install OneFlow, please refer to install-oneflow for more details.

Then install the latest stable release of flowvision

pip install flowvision==0.2.1

Overview of flowvision structure

Vision Models	Components	Augmentation and Datasets
Classification AlexNet SqueezeNet VGG GoogleNet InceptionV3 ResNet ResNeXt ResNeSt SENet DenseNet ShuffleNetV2 MobileNetV2 MobileNetV3 MNASNet Res2Net EfficientNet GhostNet RegNet ReXNet Vision Transformer DeiT PVT Swin Transformer CSwin Transformer CrossFormer PoolFormer Mlp Mixer ResMLP gMLP ConvMixer ConvNeXt LeViT RegionViT UniFormer VAN MobileViT DeiT-III CaiT DLA GENet HRNet FAN Detection SSD SSDLite Faster RCNN RetinaNet Segmentation FCN DeepLabV3 Neural Style Transfer StyleNet Face Recognition IResNet	Attention Layers SE BAM CBAM ECA Non Local Attention Global Context Gated Channel Transform Coordinate Attention Regularization Layers Drop Block Drop Path Stochastic Depth LayerNorm2D Basic Layers Patch Embedding Mlp Block FPN Activation Layers Hard Sigmoid Hard Swish Initialization Function Truncated Normal Lecun Normal LR Scheduler StepLRScheduler MultiStepLRScheduler CosineLRScheduler LinearLRScheduler PolyLRScheduler TanhLRScheduler Loss LabelSmoothingCrossEntropy SoftTargetCrossEntropy	Basic Augmentation CenterCrop RandomCrop RandomResizedCrop FiveCrop TenCrop RandomVerticalFlip RandomHorizontalFlip Resize RandomGrayscale GaussianBlur Advanced Augmentation Mixup CutMix AugMix RandomErasing Rand Augmentation Auto Augmentation Datasets CIFAR10 CIFAR100 COCO FashionMNIST ImageNet VOC

Documentation

Please refer to docs for full API documentation and tutorials

ChangeLog

Please refer to ChangeLog for details and release history

Model Zoo

We have conducted all the tests under the same setting, please refer to the model page here for more details.

Quick Start

Create a model

In flowvision we support two ways to create a model.

• Import the target model from flowvision.models, e.g., create alexnet from flowvision

from flowvision.models.alexnet import alexnet
model = alexnet()

# will download the pretrained model
model = alexnet(pretrained=True)

# customize model to fit different number of classes
model = alexnet(num_classes=100)

• Or create model in an easier way by using ModelCreator, e.g., create alexnet model by ModelCreator

from flowvision.models import ModelCreator
alexnet = ModelCreator.create_model("alexnet")

# will download the pretrained model
alexnet = ModelCreator.create_model("alexnet", pretrained=True)

# customize model to fit different number of classes
alexnet = ModelCreator.create_model("alexnet", num_classes=100)

Tabulate all models with pretrained weights

ModelCreator.model_table() returns a tabular results of available models in flowvision. To check all of pretrained models, pass in pretrained=True in ModelCreator.model_table().

from flowvision.models import ModelCreator
all_pretrained_models = ModelCreator.model_table(pretrained=True)
print(all_pretrained_models)

You can get the results like:

╒════════════════════════════════════════════╤══════════════╕
│ Supported Models                           │ Pretrained   │
╞════════════════════════════════════════════╪══════════════╡
│ alexnet                                    │ true         │
├────────────────────────────────────────────┼──────────────┤
│ convmixer_1024_20                          │ true         │
├────────────────────────────────────────────┼──────────────┤
│ convmixer_1536_20                          │ true         │
├────────────────────────────────────────────┼──────────────┤
│ convmixer_768_32_relu                      │ true         │
├────────────────────────────────────────────┼──────────────┤
│ crossformer_base_patch4_group7_224         │ true         │
├────────────────────────────────────────────┼──────────────┤
│ crossformer_large_patch4_group7_224        │ true         │
├────────────────────────────────────────────┼──────────────┤
│ crossformer_small_patch4_group7_224        │ true         │
├────────────────────────────────────────────┼──────────────┤
│ crossformer_tiny_patch4_group7_224         │ true         │
├────────────────────────────────────────────┼──────────────┤
│                    ...                     │ ...          │
├────────────────────────────────────────────┼──────────────┤
│ wide_resnet101_2                           │ true         │
├────────────────────────────────────────────┼──────────────┤
│ wide_resnet50_2                            │ true         │
╘════════════════════════════════════════════╧══════════════╛

Search for supported model by Wildcard

It is easy to search for model architectures by using Wildcard as below:

from flowvision.models import ModelCreator
all_efficientnet_models = ModelCreator.model_table("**efficientnet**")
print(all_efficientnet_models)

You can get the results like:

╒════════════════════╤══════════════╕
│ Supported Models   │ Pretrained   │
╞════════════════════╪══════════════╡
│ efficientnet_b0    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b1    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b2    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b3    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b4    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b5    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b6    │ true         │
├────────────────────┼──────────────┤
│ efficientnet_b7    │ true         │
╘════════════════════╧══════════════╛

List all models supported in flowvision

ModelCreator.model_list has similar function as ModelCreator.model_table but return a list object, which gives the user a more flexible way to check the supported model in flowvision.

• List all models with pretrained weights

from flowvision.models import ModelCreator
all_pretrained_models = ModelCreator.model_list(pretrained=True)
print(all_pretrained_models[:5])

You can get the results like:

['alexnet', 
 'convmixer_1024_20', 
 'convmixer_1536_20', 
 'convmixer_768_32_relu', 
 'crossformer_base_patch4_group7_224']

• Support wildcard search

from flowvision.models import ModelCreator
all_efficientnet_models = ModelCreator.model_list("**efficientnet**")
print(all_efficientnet_models)

You can get the results like:

['efficientnet_b0', 
 'efficientnet_b1', 
 'efficientnet_b2', 
 'efficientnet_b3', 
 'efficientnet_b4', 
 'efficientnet_b5', 
 'efficientnet_b6', 
 'efficientnet_b7']

Disclaimer on Datasets

This is a utility library that downloads and prepares public datasets. We do not host or distribute these datasets, vouch for their quality or fairness, or claim that you have license to use the dataset. It is your responsibility to determine whether you have permission to use the dataset under the dataset's license.

If you're a dataset owner and wish to update any part of it (description, citation, etc.), or do not want your dataset to be included in this library, please get in touch through a GitHub issue. Thanks for your contribution to the ML community!