Practice on cifi10 and cifar100(Vgg, ResNet, ResNeXt, Mobilenet, Mobilenetv2) ,there will be more.
This is my experiment eviroument
python3.5 pytorch1.2.0+cuda10.0 tensorboard 2.2.2(optional)
git clone https://github.com/UestcJay/cifar10_cifar100_pytorch.git
cd cifar10_cifar100_pytorch
you can use cifar100 and cifar10 dataset from torchvision directly as in the code, but most time it is not convenient for us to download the datasets for chinese, then you can download from cifar10 and cifar100, the key is kyfq. If the link fails, just feel free to open a issue to inform me. I also kept the sample code for writing your own dataset module in dataset.py. You can refer it!
pip install tensorboard
mkdir runs
Run tensorboard
tensorboard --logdir='runs' --port=6006 --host='localhost'
python train.py #train cifar100
python train_cifar10.py #train cifar10
You can change the default parameters to select the corresponding network.
sometimes, you might want to use warmup
training by set -warm to 1 or 2, to prevent network diverge during early training phase.
python test.py
it reports top-1 error and top-5 error.
If you meet any question, just feel free to open a new issue. Thanks for your attention!