pytorch-cifar-100

practice on cifar-100 using pytorch

inspired by this repo

Requirements

• python3.5
• pytorch3.1
• tensorflow1.4
• cuda8.0
• cudnnv5

Usage

1. enter directory

$ cd pytorch-cifar-100

2. run tensorbard

$ tensorboard --logdir='runs' --port=6006

3. train the model

$ python train.py

Training

I train model for 140 epoch set learning rate at:

• epoch < 60, lr = 0.1
• epoch < 100, lr = 0.01
• epoch < 140, lr = 0.001

I found that training more epoch when lr = 0.1 can improve my model prformance by %1 or %2, but add more epoch at lr = 0.01 or lr = 0.001 own't make much difference.So I decide to train my model for more epoch whye lr = 0.1

Results

dataset	network	params	top1 err	top5 err
cifar100	resnet101	42.7M	22.22	5.61
cifar100	vgg16_bn	34.0.7M	29.57	9.67