I have tried everything and unable to solve this problem. I am getting this problem on 3 different datasets I have tried. GPU is K80 . Please find the complete code in this drive link. I have pasted the dataloader , model and training step here:
class age_dataset(Dataset):
def __init__(self,label,image_dir,transform = None):
self.data = pd.read_csv(label)
self.img_dir = image_dir
self.transform = transform
self.id = self.data.iloc[:,1:2].values
def __len__(self):
return len(self.id)
def __getitem__(self,idx):
img_name = os.path.join(self.img_dir,self.id[idx,0])
image = cv2.imread(img_name)
age1 = self.data.iloc[:,2:3].values
age = age1[idx,:]
if self.transform:
image = self.transform(image)
return image,int(age)
train_transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize((224,224)),
transforms.RandomRotation(10),
transforms.ToTensor(),
transforms.Normalize([0.39125082, 0.45353994, 0.5837345 ], [0.20129617, 0.21080811, 0.24163313])
])
val_transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize([0.39125082, 0.45353994, 0.5837345 ], [0.20129617, 0.21080811, 0.24163313])
])
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
age_data = age_dataset('./agedetect/label.csv','./UTKFace' , transform = train_transform)
test_size = 0.2
valid_size = 0.1
data_len = len(age_data)
indices = list(range(data_len))
np.random.shuffle(indices)
split1 = int(np.floor(valid_size * data_len))
split2 = int(np.floor(test_size * data_len))
split2 = split2+split1
valid_idx , test_idx, train_idx = indices[:split1], indices[split1:split2] , indices[split2:]
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
test_sampler = SubsetRandomSampler(test_idx)
train_loader = DataLoader(age_data, batch_size=32, sampler=train_sampler)
valid_loader = DataLoader(age_data, batch_size=32 , sampler=valid_sampler)
test_loader = DataLoader(age_data, batch_size=32 , sampler=test_sampler)
dataloaders = {'train':train_loader,'val':valid_loader}
resnet50 = models.resnet18(pretrained=True)
num_ftrs = resnet50.fc.in_features
resnet50.fc = nn.Sequential(nn.Linear(num_ftrs,256),nn.ReLU(),nn.Linear(256,104),nn.LogSoftmax(dim=1))
resnet50 = resnet50.to(device)
criterion = nn.CrossEntropyLoss()
optimizer_ft = optim.Adam(resnet50.parameters(), lr=0.1)
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=3, gamma=0.1)
def train_model(model, criterion, optimizer, scheduler, num_epochs=10):
since = time.time()
model.to(device)
best_acc = 0.0
i = 0
for epoch in range(num_epochs):
print('Epoch:',epoch)
for phase in ['train', 'val']:
if phase == ' train':
scheduler.step()
model.train()
else:
model.eval()
running_loss = 0.0
running_corrects = 0
total = 0
for inputs, labels in dataloaders[phase]:
labels = labels.to(device)
#labels = labels.type(torch.cuda.FloatTensor)
inputs = inputs.view(inputs.shape[0],3,224,224)
inputs = inputs.to(device)
inputs = inputs.type(torch.cuda.FloatTensor)
# zero the parameter gradients
optimizer.zero_grad()
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
#preds = preds.reshape(preds.size(0),-1)
loss = criterion(outputs, labels)
# print(preds)
if phase == 'train':
loss.backward()
optimizer.step()
# statistics
#running_loss += loss.item() * inputs.size(0)
#print(labels.shape)
a = preds.data
#running_corrects = running_corrects + torch.sum(preds == labels.data)
#total += labels.size(0)
i = i+1
'''if(i%100==0):
print('loss :{} accuracy : {}'.format(running_loss/(i*32),running_corrects.double()/(total)))
#print(i)
epoch_loss = running_loss/(len(dataloaders[phase])*32)
epoch_acc = running_corrects.double()/(len(dataloaders[phase])*32)
'''
#writer.add_scalar('./sanchit/scalar1', epoch_loss , epoch)
print('{} Loss: {:.4f} , acc: {:.4f}'.format(phase, epoch_loss , epoch_acc))
time_elapsed = time.time() - since
print('Training complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
best_acc = epoch_acc
print('Best val Acc: {:4f}'.format(best_acc))
torch.save(resnet50, '/content/genreweights-{}.h5'.format(epoch_acc))
# load best model weights
#model.load_state_dict(best_model_wts)
return model
model_ft = train_model(resnet50, criterion, optimizer_ft, exp_lr_scheduler,num_epochs=5)
Thanks in advance