unsky / Mixup
mixup: Beyond Empirical Risk Minimization
Stars: ✭ 96
Programming Languages
python
139335 projects - #7 most used programming language
Labels
Projects that are alternatives of or similar to Mixup
Competitive Inner Imaging Senet
Source code of paper: (not available now)
Stars: ✭ 89 (-7.29%)
Mutual labels: mxnet
Aws Machine Learning University Accelerated Cv
Machine Learning University: Accelerated Computer Vision Class
Stars: ✭ 1,068 (+1012.5%)
Mutual labels: mxnet
Mxnet Gluon Syncbn
MXNet Gluon Synchronized Batch Normalization Preview
Stars: ✭ 78 (-18.75%)
Mutual labels: mxnet
Vapoursynthcolab
AI Video Processing/Upscaling With VapourSynth in Google Colab
Stars: ✭ 47 (-51.04%)
Mutual labels: mxnet
Mnasnet Pretrained Model
An Implementation of Mnasnet with Pretrained Model
Stars: ✭ 94 (-2.08%)
Mutual labels: mxnet
Insightface
State-of-the-art 2D and 3D Face Analysis Project
Stars: ✭ 10,886 (+11239.58%)
Mutual labels: mxnet
Ko en neural machine translation
Korean English NMT(Neural Machine Translation) with Gluon
Stars: ✭ 55 (-42.71%)
Mutual labels: mxnet
Audio Pretrained Model
A collection of Audio and Speech pre-trained models.
Stars: ✭ 61 (-36.46%)
Mutual labels: mxnet
Mxnet Oneclick
use mxnet to train your own data with just oneclick
Stars: ✭ 81 (-15.62%)
Mutual labels: mxnet
Go Mxnet Predictor
go binding for mxnet c_predict_api to do inference with pre-trained model
Stars: ✭ 52 (-45.83%)
Mutual labels: mxnet
Learn Convolutional Neural Network For Face Anti Spoofing
Implementation of "Learn Convolutional Neural Network for Face Anti-Spoofing" paper
Stars: ✭ 90 (-6.25%)
Mutual labels: mxnet
Gluon2pytorch
Gluon to PyTorch deep neural network model converter
Stars: ✭ 70 (-27.08%)
Mutual labels: mxnet
MixUp
This is an implement and Improvement on mixup: Beyond Empirical Risk Minimization https://arxiv.org/abs/1710.09412
The improvement
- add backward
- add mix rate
Two scenes:
The detail design of MixUp layer:
The results:
The symbol of resnet50 is writen by mxnet https://github.com/apache/incubator-mxnet/tree/master/example/image-classification/symbols, there have many versions. And i havenot do any optimizion for it. All the results are based on this baseline.
| cifar10 | alpha | mix_rate | test Acc | initial learning rate | batch size |
|---|---|---|---|---|---|
| (ERM)resnet50 90epoch | - | - | 0.87900390625 | 0.05 | 256 |
| (ERM)resnet50 200epoch | - | - | 0.89365234375 | 0.05 | 256 |
| (ERM)resnet50 300epoch | - | - | 0.8931640625 | 0.05 | 256 |
| (mixup)resnet50 90epoch | 0.2 | 0.7 | 0.8609375 | 0.7 | 256 |
| (mixup)resnet50 200epoch | 0.2 | 0.7 | 0.91611328125 | 0.7 | 256 |
| (mixup)resnet50 300epoch | 0.2 | 0.7 | 0.9224609375 | 0.7 | 256 |
| mixup in feature maps（resnet50 head conv）90epoch | 0.2 | 0.7 | 0.8544921875 | 0.7 | 256 |
| mixup in feature maps（resnet50 head conv）200epoch | 0.2 | 0.7 | 0.91796875 | 0.7 | 256 |
| mixup in feature maps（resnet50 head conv）300epoch | 0.2 | 0.7 | 0.91845703125 | 0.7 | 256 |
MixUp
Mixup in feature map （resnet50 head conv）
ERM
Usage
install mxnet0.12 The mixup is in:symbols/mixup.py you can use it in your codes like:
data ,label = mx.sym.Custom(data= data,label = label,alpha = 0.2,num_classes = num_classes,batch_size = batch_size,mix_rate =0.7,op_type = 'MixUp')
label is the vector like [4,8,...9]
download the dataset
http://data.mxnet.io/data/cifar10/cifar10_val.rec
http://data.mxnet.io/data/cifar10/cifar10_train.rec
train & test:
./train.sh
./test.sh
Reference
Zhang H, Cisse M, Dauphin Y N, et al. mixup: Beyond Empirical Risk Minimization[J]. arXiv preprint arXiv:1710.09412, 2017.
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].