Skip to content

frotms/image_classification_mxnet_gluon

Repository files navigation

image-classification-mxnet

This repo is designed for those who want to start their projects of image classification. It provides fast experiment setup and attempts to maximize the number of projects killed within the given time. It includes a few Convolutional Neural Network modules.You can build your own dnn easily.

Requirements

Python3 support only. Tested on CUDA9.0, cudnn7.

  • albumentations==0.1.1
  • easydict==1.8
  • imgaug==0.2.6
  • opencv-python==3.4.3.18
  • protobuf==3.6.1
  • scikit-image==0.14.0
  • mxboard 0.1.0
  • mxnet-cu90 1.3.0.post0

model

net inputsize
vggnet 224
alexnet 224
resnet 224
inceptionV3 299
squeezenet 224
densenet 224
mobilenet 224
nasnet 331
resnext 224
senet 224
se_resnet 224
squeezenet 224
... ...

pre-trained model

you can download pretrain model with url in ($net-module.py)
And the accuracy of ImageNet pre-trained models is illustrated in the following URLs:

From mxnet-gluon-vision package:

  • ResNet (resnet18_v1, resnet34_v1, resnet50_v1, resnet101_v1, resnet152_v1, resnet18_v2, resnet34_v2, resnet50_v2, resnet101_v2, resnet152_v2)
  • DenseNet (densenet121, densenet169, densenet201, densenet161)
  • Inception v3 (inception_v3)
  • VGG (vgg11, vgg11_bn, vgg13, vgg13_bn, vgg16, vgg16_bn, vgg19, vgg19_bn)
  • SqueezeNet (squeezenet1_0, squeezenet1_1)
  • AlexNet (alexnet)
  • Mobilenet(mobilenet1_0, mobilenet_v2_1_0, mobilenet0_75, mobilenet_v2_0_75, mobilenet0_5, mobilenet_v2_0_5, mobilenet0_25, mobilenet_v2_0_25)
  • ResNeXt (resnext50_32x4d, resnext101_32x4d, resnext101_64x4d, se_resnext50_32x4d, se_resnext101_32x4d, se_resnext101_64x4d)
  • NASNet (nasnet_4_1056, nasnet_5_1538, nasnet_7_1920, nasnet_6_4032)
  • SENet (senet_52, senet_103, senet_154)
  • SE_ResNeXt (se_resnet18_v1, se_resnet34_v1, se_resnet50_v1, se_resnet101_v1, se_resnet152_v1, se_resnet18_v2, se_resnet34_v2, se_resnet50_v2, se_resnet101_v2, se_resnet152_v2)

usage

configuration

configure description
model_module_name eg: vgg_module
model_net_name net function name in module, eg:vgg16
gpu_id eg: single GPU: "0", multi-GPUs:"0,1,3,4,7"
is_mxboard if use tensorboard for visualization
evaluate_before_train evaluate accuracy before training
shuffle shuffle your training data
data_aug augment your training data
model_auto_download "True" for download pretrained model with gluonAPI in a automatic manner
img_height input height
img_width input width
num_channels input channel
num_classes output number of classes
batch_size train batch size
dataloader_workers number of workers when loading data
learning_rate learning rate
learning_rate_decay learning rate decat rate
learning_rate_decay_epoch learning rate decay per n-epoch
train_mode eg: "fromscratch","finetune","update"
file_label_separator separator between data-name and label. eg:"----"
pretrained_path pretrain model path
pretrained_file pretrain model name. eg:"alexnet-owt-4df8aa71.pth"
pretrained_model_num_classes output number of classes when pretrain model trained. eg:1000 in imagenet
save_path model path when saving
save_name model name when saving
train_data_root_dir training data root dir
val_data_root_dir testing data root dir
train_data_file a txt filename which has training data and label list
val_data_file a txt filename which has testing data and label list

Training

1.make your training &. testing data and label list with txt file:

txt file with single label index eg:

apple.jpg----0
k.jpg----3
30.jpg----0
data/2.jpg----1
abc.jpg----1

2.configuration

3.train

python3 train.py

Inference

python3 inference.py --image images/image_00002.jpg --module vgg --net vgg16 --model /hdd/datasets/flower_102/save_model/model.params --size 224 --cls 102

mxboard

tensorboard --logdir=./logs/ 

logdir is log dir in your project dir

experiment

There is integrated with the project using mxboard library which porved to be very useful as there is no official visualization library in mxnet. There is the learning curves for the flower-102 dataset experiment(top-1 acc: 87.74% for simple experiment).


top-5:
passion flower: 100.0%
king protea: 3.3845638097718123e-09%
barbeton daisy: 3.775002879735645e-10%
purple coneflower: 1.621601507587056e-10%
spring crocus: 1.3393154857724299e-10%

References

1.http://mxnet.incubator.apache.org/
2.https://github.com/dmlc/gluon-cv/tree/master/gluoncv/model_zoo
3.https://github.com/chinakook/Awesome-MXNet
4.https://github.com/awslabs/mxboard
5.http://www.robots.ox.ac.uk/~vgg/data/flowers/102
6.https://github.com/victoresque/pytorch-template

About

image classification for mxnet-gluon refer to pytorch-project-template, train a model of classification easily by modifying a json configuration

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages