train.py

"""
 Copyright (c) 2020 Intel Corporation
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
 You may obtain a copy of the License at
      http://www.apache.org/licenses/LICENSE-2.0
 Unless required by applicable law or agreed to in writing, software
 distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
"""

import torch
import torch.optim as optim
import torchvision.transforms as transforms
import argparse
import torch.nn as nn
import torch.nn.functional as F
from utils.dataset_eye import EyeDataset
import torch.onnx
import os.path


class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 10, kernel_size=3)
        self.conv2 = nn.Conv2d(10, 20, kernel_size=3)
        self.conv3 = nn.Conv2d(20, 50, kernel_size=3)
        self.conv4 = nn.Conv2d(50, 2, kernel_size=1, bias=False, padding=0, stride=1)
        self.max_pool2d = nn.MaxPool2d((4, 4))
        self.softmax = nn.Softmax(dim=1)

    def forward(self, x):
        x = F.relu(F.max_pool2d(self.conv1(x), 2))
        x = F.relu(F.max_pool2d(self.conv2(x), 2))
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.max_pool2d(x)
        x = self.softmax(x)
        return x


def parse_args():
    parser = argparse.ArgumentParser(description='Eye state classifier')
    parser.add_argument('-d', '--data_root', required=True, help='Path to dataset')
    parser.add_argument('-e', '--epoch', default=20, type=int, help='Number of epochs')
    parser.add_argument('-b', '--batch_size', default=124, type=int)
    parser.add_argument('-l', '--learning_rate', default=0.1, type=float, help='Star learning rate')
    parser.add_argument('-p', '--pretrained', help='Path to pretrained weights')
    parser.add_argument('-m', '--model_dir', default='model', help='Path to save snapshots')
    args = parser.parse_args()
    return args


def train(net, loader, device, optimizer):
    running_loss = 0.0
    criterion = nn.CrossEntropyLoss()
    for i, (inputs, labels, _) in enumerate(loader):
        inputs = inputs.to(device)
        labels = labels.to(device)
        optimizer.zero_grad()
        outputs = net(inputs).view(len(labels), 2)
       
        loss = criterion(outputs, labels)
        
        loss.backward()
        optimizer.step()
        running_loss += loss
        # if i % 1 == 9:    # print every 20 mini-batches
        print('iter: %5d loss: %.3f' % (i + 1, running_loss / 10))
        running_loss = 0.0


def test(net, loader, device):
    corrected = 0.
    total = 0.
    with torch.no_grad():
        for i, (inputs, labels, fname) in enumerate(loader):           
            inputs = inputs.to(device)
            labels = labels.to(device)
            outputs = net(inputs)
            _, predicted = torch.max(outputs, 1)
            
            for label, pred in zip(labels, predicted):
                if label == pred[0][0]:
                    corrected += 1 
                total += 1

    print("Test accuracy: {}".format(corrected / total))


def main():
    args = parse_args()
    torch.manual_seed(100)
    device = torch.device("cuda")
    net = Net().to(device)
    if args.pretrained:
        net.load_state_dict(torch.load(args.pretrained))

    lr = args.learning_rate
    optimizer = optim.SGD(net.parameters(), lr=lr, momentum=0.9)

    train_transforms = transforms.Compose([
                        transforms.Resize((32, 32)),
                        transforms.RandomHorizontalFlip(),
                        transforms.ToTensor(),
                        transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[1, 1, 1], inplace=False),
                        ])
    test_transforms = transforms.Compose([
                        transforms.Resize((32, 32)),
                        transforms.ToTensor(),
                        transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[1, 1, 1], inplace=False),
                        ])

    train_db = EyeDataset(args.data_root, 'train', train_transforms)
    test_db = EyeDataset(args.data_root, 'val', test_transforms)

    train_db_loader = torch.utils.data.DataLoader(train_db, batch_size=args.batch_size, shuffle=True, num_workers=5)
    test_db_loader = torch.utils.data.DataLoader(test_db, batch_size=1, num_workers=1)

    if not os.path.exists(args.model_dir):
        os.mkdir(args.model_dir)

    for epoch in range(args.epoch):         
        if epoch in [10, 15, 25]:
            lr = 0.1 * lr
        for param_group in optimizer.param_groups:
            param_group['lr'] = lr
        print ('epoch={}, lr={}'.format(epoch, lr))
        train(net, train_db_loader, device, optimizer)
        test(net, test_db_loader, device)

        snap_path = os.path.join(args.model_dir, "open_closed_eye_epoch_{}.pth".format(epoch))
        onnx_path = os.path.join(args.model_dir, "open_closed_eye_epoch_{}.onnx".format(epoch))
        torch.save(net.state_dict(), snap_path)
        dummy_input = torch.randn(1, 3, 32, 32, requires_grad=False).to(device)
        torch.onnx.export(net, dummy_input, onnx_path, export_params=True)


if __name__ == '__main__':
    main()