''' https://github.com/kuangliu/pytorch-cifar ResNet in PyTorch. For Pre-activation ResNet, see 'preact_resnet.py'. Reference: [1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun Deep Residual Learning for Image Recognition. arXiv:1512.03385 ''' import torch from torch import nn from torch.nn import functional as F from torch_lr_finder import LRFinder class BasicBlock(nn.Module): expansion = 1 def __init__(self, in_planes, planes, stride=1): super(BasicBlock, self).__init__() self.conv1 = nn.Conv2d( in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(planes) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(planes) self.shortcut = nn.Sequential() if stride != 1 or in_planes != self.expansion*planes: self.shortcut = nn.Sequential( nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(self.expansion*planes) ) def forward(self, x): out = F.relu(self.bn1(self.conv1(x))) out = self.bn2(self.conv2(out)) out += self.shortcut(x) out = F.relu(out) return out class ResNet(nn.Module): def __init__(self, block, num_blocks, num_classes=10): super(ResNet, self).__init__() self.in_planes = 64 self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(64) self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1) self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2) self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2) self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2) self.linear = nn.Linear(512*block.expansion, num_classes) def _make_layer(self, block, planes, num_blocks, stride): strides = [stride] + [1]*(num_blocks-1) layers = [] for stride in strides: layers.append(block(self.in_planes, planes, stride)) self.in_planes = planes * block.expansion return nn.Sequential(*layers) def forward(self, x): out = F.relu(self.bn1(self.conv1(x))) out = self.layer1(out) out = self.layer2(out) out = self.layer3(out) out = self.layer4(out) out = F.avg_pool2d(out, 4) out = out.view(out.size(0), -1) out = self.linear(out) return out def ResNet18(): return ResNet(BasicBlock, [2, 2, 2, 2]) import torch.nn as nn from torch.optim.lr_scheduler import OneCycleLR from torch.utils.data import DataLoader import matplotlib.pyplot as plt from data_loader import CifarAlbumentationsDataset,\ CIFAR_CLASS_LABELS, TRAIN_TRANSFORM, TEST_TRANSFORM import model from torch_lr_finder import LRFinder import torch import torch.nn as nn import torch.nn.functional as F from pytorch_lightning import LightningModule from torch.optim.lr_scheduler import OneCycleLR from torchmetrics.functional import accuracy class LitResnet(LightningModule): def __init__(self, lr=0.03, batch_size=512): super().__init__() self.save_hyperparameters() self.criterion = nn.CrossEntropyLoss() self.model = ResNet18() def forward(self, x): return self.model(x) def training_step(self, batch, batch_idx): x, y = batch output = self.forward(x) loss = self.criterion(output, y) self.log("train_loss", loss) acc = accuracy(torch.argmax(output, dim=1), y, 'multiclass', num_classes=10) self.log(f"train_acc", acc, prog_bar=True) return loss def evaluate(self, batch, stage=None): x, y = batch output = self.forward(x) loss = self.criterion(output, y) preds = torch.argmax(output, dim=1) acc = accuracy(preds, y, 'multiclass', num_classes=10) if stage: self.log(f"{stage}_loss", loss, prog_bar=True) self.log(f"{stage}_acc", acc, prog_bar=True) def validation_step(self, batch, batch_idx): self.evaluate(batch, "val") def test_step(self, batch, batch_idx): self.evaluate(batch, "test") # todo # change the default for num_iter def lr_finder(self, optimizer, num_iter=200,): lr_finder = LRFinder(self, optimizer, self.criterion, device=self.device) lr_finder.range_test( self.train_dataloader(), end_lr=1, num_iter=num_iter, step_mode='exp', ) ax, suggested_lr = lr_finder.plot(suggest_lr=True) # todo # how to log maplotlib images # self.logger.experiment.add_image('lr_finder', plt.gcf(), 0) lr_finder.reset() return suggested_lr def configure_optimizers(self): optimizer = torch.optim.SGD( self.parameters(), lr=self.hparams.lr, momentum=0.9, weight_decay=5e-4, ) suggested_lr = self.lr_finder(optimizer) steps_per_epoch = len(self.train_dataloader()) scheduler_dict = { "scheduler": OneCycleLR( optimizer, max_lr=suggested_lr, steps_per_epoch=steps_per_epoch, epochs=self.trainer.max_epochs, pct_start=5/self.trainer.max_epochs, three_phase=False, div_factor=100, final_div_factor=100, anneal_strategy='linear', ), "interval": "step", } return {"optimizer": optimizer, "lr_scheduler": scheduler_dict} #################### # DATA RELATED HOOKS #################### def prepare_data(self, data_path='../data'): CifarAlbumentationsDataset( data_path, train=True, download=True) CifarAlbumentationsDataset( data_path, train=False, download=True) def setup(self, stage=None, data_dir='../data'): if stage == "fit" or stage is None: self.train_dataset = CifarAlbumentationsDataset(data_dir, train=True, transform=TRAIN_TRANSFORM) self.test_dataset = CifarAlbumentationsDataset(data_dir, train=False, transform=TEST_TRANSFORM) def train_dataloader(self): return DataLoader(self.train_dataset, batch_size=self.hparams.batch_size, shuffle=True, pin_memory=True) #num_workers=4, def val_dataloader(self): return DataLoader(self.test_dataset, batch_size=self.hparams.batch_size, shuffle=False, pin_memory=True)