我正在建立一个CNN，以便对LFW数据集进行图像分类。
为此，我具有以下数据集：
class LFW(object):
def __init__(self, root, specific_folder, img_extension, preprocessing_method=None, crop_size=(96, 112)):
"""
Dataloader of the LFW dataset.
root: path to the dataset to be used.
specific_folder: specific folder inside the same dataset.
img_extension: extension of the dataset images.
preprocessing_method: string with the name of the preprocessing method.
crop_size: retrieval network specific crop size.
"""
self.preprocessing_method = preprocessing_method
self.crop_size = crop_size
self.imgl_list = []
self.classes = []
self.people = []
self.model_align = None
# read the file with the names and the number of images of each people in the data set
with open(os.path.join(root, 'people.txt')) as f:
people = f.read().splitlines()[1:]
# get only the people that have more than 20 images
for p in people:
p = p.split('\t')
if len(p) > 1:
if int(p[1]) >= 20:
for num_img in range(1, int(p[1]) + 1):
self.imgl_list.append(os.path.join(root, specific_folder, p[0], p[0] + '_' +
'{:04}'.format(num_img) + '.' + img_extension))
self.classes.append(p[0])
self.people.append(p[0])
le = preprocessing.LabelEncoder()
self.classes = le.fit_transform(self.classes)
print(len(self.imgl_list), len(self.classes), len(self.people))
def __getitem__(self, index):
imgl = imageio.imread(self.imgl_list[index])
cl = self.classes[index]
# if image is grayscale, transform into rgb by repeating the image 3 times
if len(imgl.shape) == 2:
imgl = np.stack([imgl] * 3, 2)
imgl, bb = preprocess(imgl, self.preprocessing_method, crop_size=self.crop_size,
is_processing_dataset=True, return_only_largest_bb=True, execute_default=True)
# append image with its reverse
imglist = [imgl, imgl[:, ::-1, :]]
# normalization
for i in range(len(imglist)):
imglist[i] = (imglist[i] - 127.5) / 128.0
imglist[i] = imglist[i].transpose(2, 0, 1)
imgs = [torch.from_numpy(i).float() for i in imglist]
return imgs, cl, imgl, bb, self.imgl_list[index], self.people[index]
def __len__(self):
return len(self.imgl_list)
为了训练我的模块，我使用以下代码
if __name__ == '__main__':
#set_trace()
args = edict({
'operation' : 'train',
'feature_file' : None,
'result_sample_path' : None,
'gpu' : 'GPU',
'path_image_query' : None,
'query_label' : 'Query label',
'dataset' : None,
'specific_dataset_folder_name' : 'lfw',
'img_extension' : 'jpg',
'preprocessing_method' : 'sphereface',
'model_name' : 'mobiface',
'batch_size' : 3,
'image_query':'/content/drive/My Drive/recfaces13/recfaces/datasets/LFW',
'train':True,
'device':'cuda'
}） 打印（参数）
# selecting the size of the crop based on the network
if args.model_name == 'mobilefacenet' or args.model_name == 'sphereface':
crop_size = (96, 112)
elif args.model_name == 'mobiface' or args.model_name == 'shufflefacenet':
crop_size = (112, 112)
elif args.model_name == 'openface':
crop_size = (96, 96)
elif args.model_name == 'facenet':
crop_size = (160, 160)
else:
raise NotImplementedError("Model " + args.model_name + " not implemented")
if args.dataset is not None:
# process whole dataset
assert args.specific_dataset_folder_name is not None, 'To process a dataset, ' \
'the flag --specific_dataset_folder_name is required.'
process_dataset(args.operation, args.model_name, args.batch_size,
args.dataset, args.specific_dataset_folder_name,
args.img_extension, args.preprocessing_method, crop_size,
args.result_sample_path, args.feature_file)
#elif args.image_query is not None:
# process unique image
# dataset = ImageDataLoader(args.image_query, args.preprocessing_method,
# crop_size, args.operation == 'extract_features')
# dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=False, num_workers=2, drop_last=False)
# features = None
elif args.operation == 'train':
##########set_trace()
net = load_net('mobilefacenet', 'gpu')
net = net.cuda()
model_name=args.model_name
dataset = LFW(args.image_query,args.specific_dataset_folder_name, args.img_extension, args.preprocessing_method, crop_size)
dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=False, num_workers=2, drop_last=False)
# data_counter_per_class = np.zeros((len(dataloader)))
# for i in range(len(dataloader)):
# path = os.path.join('image_query', dataloader[i])
# files = get_files_from_folder(path)
# data_counter_per_class[i] = len(files)
# test_counter = np.round(data_counter_per_class * (1 - train_ratio))
#dataloader1=dataloader.split(',')
#train,test=train_test_split(dataloader,test_size=0.2)
#trainloader = torch.utils.data.DataLoader(dataset, batch_size=4, shuffle=True, num_workers=2, drop_last=False)
# testloader = torch.utils.data.DataLoader(dataset, batch_size=4, shuffle=False, num_workers=2, drop_last=False) //create path//
#create array of data path split that data path and
features = None
if args.feature_file is not None and os.path.isfile(args.feature_file):
features = scipy.io.loadmat(args.feature_file)
epoch = 2
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
train_loss = list()
#set_trace()
for i, data in enumerate(dataloader):
inpss, labs = data
for inps in inpss:
#set_trace()
inps, labs = inps.cuda(args['device']), labs.cuda(args['device'])
inps.squeeze_(0)
labs.squeeze_(0)
inps = Variable(inps).cuda(args['device'])
labs = Variable(labs).cuda(args['device'])
optimizer.zero_grad()
#set_trace()
outs = net(inps)
soft_outs = F.softmax(outs, dim=1)
prds = soft_outs.data.max(1)[1]
loss = criterion(outs, labs)
loss.backward()
optimizer.step()
prds = prds.squeeze_(1).squeeze_(0).cpu().numpy()
inps_np = inps.detach().squeeze(0).cpu().numpy()
labs_np = labs.detach().squeeze(0).cpu().numpy()
train_loss.append(loss.data.item())
print('[epoch %d], [iter %d / %d], [train loss %.5f]' % (epoch, i + 1, len(train_loader), np.asarray(train_loss).mean()))
错误弹出，我有3D但需要4D我不确定如何解决。我现在是否需要更改数据负载或模型或训练非常困难，现在有人可以帮忙