Convolutional neural networks for computer vision
This repo is used to research convolutional networks for task of computer vision. For this purpose, the repo contains
(re)implementations of various classification and segmentation models and scripts for training/evaluating/converting.
The following frameworks are used:
For each supported framework, there is a PIP-package containing pure models without auxiliary scripts. List of packages:
Currently, models are mostly implemented on Gluon and then ported to other frameworks. Some models are pretrained on
ImageNet-1K, CIFAR-10/100,
SVHN, CUB-200-2011,
Pascal VOC2012, ADE20K,
Cityscapes, and COCO datasets. All pretrained weights
are loaded automatically during use. See examples of such automatic loading of weights in the corresponding sections of
the documentation dedicated to a particular package:
Installation
To use training/evaluating scripts as well as all models, you need to clone the repository and install dependencies:
git clone [email protected]:osmr/imgclsmob.git
pip install -r requirements.txt
Table of implemented classification models
Some remarks:
Repo
is an author repository, if it exists.
A
, B
, C
, D
, and E
means the implementation of a model for ImageNet-1K, CIFAR-10, CIFAR-100, SVHN, and CUB-200-2011, respectively.
A+
, B+
, C+
, D+
, and E+
means having a pre-trained model for corresponding datasets.
Model
Gluon
PyTorch
Chainer
Keras
TF
Paper
Repo
Year
AlexNet
A+
A+
A+
A+
A+
link
link
2012
ZFNet
A
A
A
A
A
link
-
2013
VGG
A+
A+
A+
A+
A+
link
-
2014
BN-VGG
A+
A+
A+
A+
A+
link
-
2015
BN-Inception
A+
A+
A+
-
-
link
-
2015
ResNet
A+B+C+D+E+
A+B+C+D+E+
A+B+C+D+E+
A+
A+
link
link
2015
PreResNet
A+B+C+D+
A+B+C+D+
A+B+C+D+
A+
A+
link
link
2016
ResNeXt
A+B+C+D+
A+B+C+D+
A+B+C+D+
A+
A+
link
link
2016
SENet
A+
A+
A+
A+
A+
link
link
2017
SE-ResNet
A+B+C+D+E+
A+B+C+D+E+
A+B+C+D+E+
A+
A+
link
link
2017
SE-PreResNet
A+B+C+D+
A+B+C+D+
A+B+C+D+
A+
A+
link
link
2017
SE-ResNeXt
A+
A+
A+
A+
A+
link
link
2017
IBN-ResNet
A+
A+
-
-
-
link
link
2018
IBN-ResNeXt
A+
A+
-
-
-
link
link
2018
IBN-DenseNet
A+
A+
-
-
-
link
link
2018
AirNet
A+
A+
A+
-
-
link
link
2018
AirNeXt
A+
A+
A+
-
-
link
link
2018
BAM-ResNet
A+
A+
A+
-
-
link
link
2018
CBAM-ResNet
A+
A+
A+
-
-
link
link
2018
ResAttNet
A
A
A
-
-
link
link
2017
SKNet
A
A
A
-
-
link
link
2019
DIA-ResNet
AB+C+D+
AB+C+D+
AB+C+D+
-
-
link
link
2019
DIA-PreResNet
AB+C+D+
AB+C+D+
AB+C+D+
-
-
link
link
2019
PyramidNet
A+B+C+D+
A+B+C+D+
A+B+C+D+
-
-
link
link
2016
DiracNetV2
A+
A+
A+
-
-
link
link
2017
ShaResNet
A
A
A
-
-
link
link
2017
CRU-Net
A+
-
-
-
-
link
link
2018
DenseNet
A+B+C+D+
A+B+C+D+
A+B+C+D+
A+
A+
link
link
2016
CondenseNet
A+
A+
A+
-
-
link
link
2017
SparseNet
A
A
A
-
-
link
link
2018
PeleeNet
A+
A+
A+
-
-
link
link
2018
Oct-ResNet
ABCD
A
A
-
-
link
-
2019
Res2Net
A
-
-
-
-
link
-
2019
WRN
A+B+C+D+
A+B+C+D+
A+B+C+D+
-
-
link
link
2016
WRN-1bit
B+C+D+
B+C+D+
B+C+D+
-
-
link
link
2018
DRN-C
A+
A+
A+
-
-
link
link
2017
DRN-D
A+
A+
A+
-
-
link
link
2017
DPN
A+
A+
A+
-
-
link
link
2017
DarkNet Ref
A+
A+
A+
A+
A+
link
link
-
DarkNet Tiny
A+
A+
A+
A+
A+
link
link
-
DarkNet-19
A
A
A
A
A
link
link
-
DarkNet-53
A+
A+
A+
A+
A+
link
link
2018
ChannelNet
A
A
A
-
A
link
link
2018
iSQRT-COV-ResNet
A
A
-
-
-
link
link
2017
RevNet
-
A
-
-
-
link
link
2017
i-RevNet
A+
A+
A+
-
-
link
link
2018
BagNet
A+
A+
A+
-
-
link
link
2019
DLA
A+
A+
A+
-
-
link
link
2017
MSDNet
A
AB
-
-
-
link
link
2017
FishNet
A+
A+
A+
-
-
link
link
2018
ESPNetv2
A+
A+
A+
-
-
link
link
2018
X-DenseNet
AB+C+D+
AB+C+D+
AB+C+D+
-
-
link
link
2017
SqueezeNet
A+
A+
A+
A+
A+
link
link
2016
SqueezeResNet
A+
A+
A+
A+
A+
link
-
2016
SqueezeNext
A+
A+
A+
A+
A+
link
link
2018
ShuffleNet
A+
A+
A+
A+
A+
link
-
2017
ShuffleNetV2
A+
A+
A+
A+
A+
link
-
2018
MENet
A+
A+
A+
A+
A+
link
link
2018
MobileNet
A+E+
A+E+
A+E+
A+
A+
link
link
2017
FD-MobileNet
A+
A+
A+
A+
A+
link
link
2018
MobileNetV2
A+
A+
A+
A+
A+
link
link
2018
MobileNetV3
A
A
A
-
-
link
-
2019
IGCV3
A+
A+
A+
A+
A+
link
link
2018
MnasNet
A+
A+
A+
A+
A+
link
-
2018
DARTS
A+
A+
A+
-
-
link
link
2018
ProxylessNAS
A+E+
A+E+
A+E+
-
-
link
link
2018
Xception
A+
A+
A+
-
-
link
link
2016
InceptionV3
A+
A+
A+
-
-
link
link
2015
InceptionV4
A+
A+
A+
-
-
link
link
2016
InceptionResNetV2
A+
A+
A+
-
-
link
link
2016
PolyNet
A+
A+
A+
-
-
link
link
2016
NASNet-Large
A+
A+
A+
-
-
link
link
2017
NASNet-Mobile
A+
A+
A+
-
-
link
link
2017
PNASNet-Large
A+
A+
A+
-
-
link
link
2017
EfficientNet
A+
A+
A+
A+
-
link
link
2019
NIN
B+C+D+
B+C+D+
B+C+D+
-
-
link
link
2013
RoR-3
B+C+D+
B+C+D+
B+C+D+
-
-
link
-
2016
RiR
B+C+D+
B+C+D+
B+C+D+
-
-
link
-
2016
ResDrop-ResNet
BCD
BCD
BCD
-
-
link
link
2016
Shake-Shake-ResNet
B+C+D+
B+C+D+
B+C+D+
-
-
link
link
2017
ShakeDrop-ResNet
BCD
BCD
BCD
-
-
link
-
2018
FractalNet
BC
BC
-
-
-
link
link
2016
NTS-Net
E+
E+
E+
-
-
link
link
2018
Table of implemented segmentation models
Some remarks:
A
corresponds to Pascal VOC2012.
B
corresponds to Pascal ADE20K.
C
corresponds to Pascal Cityscapes.
D
corresponds to Pascal COCO.
Model
Gluon
PyTorch
Chainer
Keras
TensorFlow
Paper
Repo
Year
PSPNet
A+B+C+D+
A+B+C+D+
A+B+C+D+
-
-
link
-
2016
DeepLabv3
A+B+CD+
A+B+CD+
A+B+CD+
-
-
link
-
2017
FCN-8s(d)
A+B+CD+
A+B+CD+
A+B+CD+
-
-
link
-
2014