def evaluate_classification(path):
W = 1280
H = 180
dataset_name = 'albedo'
transform = tv.transforms.Compose([
Crop(H),
tv.transforms.ToTensor(),
tv.transforms.Normalize(mean=[0.5], std=[1.0])
])
dataset = build_datasets(path, dataset_name, transform)
loader = DataLoader(dataset,
batch_size=len(dataset),
shuffle=False,
num_workers=4)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
loaded_dict = torch.load('model.m', map_location='cpu')
print('Loading model...')
model = Alexnet()
model.load_state_dict(loaded_dict['model'])
model = model.to(device)
model.eval()
with torch.no_grad():
print('Loading data...')
data = next(iter(loader))
X = data['image'].float().to(device)
X = X.expand(-1, 3, -1, -1)
print('Evaluating model...')
pred = model(X).round().byte().view(-1)
print('Writing to file...')
pred_str = [
build_string(name, value)
for name, value in zip(data['image_name'], pred.tolist())
]
with open("classification_results.txt", "a") as file:
file.write(''.join(pred_str))
return pred
path_training_ok = '/mnt/DATA/beantech_contestAI/Dataset2/campioni OK'
path_training_ko = '/mnt/DATA/beantech_contestAI/Dataset2/campioni KO'
path_validation_ok = '/mnt/DATA/beantech_contestAI/Dataset1/campioni OK'
path_validation_ko = '/mnt/DATA/beantech_contestAI/Dataset1/campioni KO'
num_epochs = 1
dataset_name = 'albedo'
tranform = tv.transforms.Compose([Crop(H), tv.transforms.ToTensor()])
# (training_set_ok,), (training_set_ko,) = preproc.build_datasets(path_training_ok, path_training_ko, dataset_name, tranform, split=True)
# training_loader_ok = DataLoader(training_set_ok, batch_size=int(opt['batch_size']/2), shuffle=True)
# training_loader_ko = DataLoader(training_set_ko, batch_size=int(opt['batch_size']/2), shuffle=True)
validation_set, = preproc.build_datasets(path_validation_ok,
path_validation_ko, dataset_name,
tranform)
val, test = preproc.split(validation_set, 0.7, random_state=55)
val_loader = DataLoader(val, batch_size=len(val), shuffle=False)
test_loader = DataLoader(test, batch_size=len(test), shuffle=False)
val_data = next(iter(val_loader))
del val_loader, val
# test_data = next(iter(test_loader))
# del test_loader, test
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#%%
H = 180
path_training_ok = join(opt['train_path'], 'campioni OK')
path_training_ko = join(opt['train_path'], 'campioni KO')
path_validation_ok = join(opt['validation_path'], 'campioni OK')
path_validation_ko = join(opt['validation_path'], 'campioni KO')
num_epochs = 1
dataset_name = 'albedo'
transform = tv.transforms.Compose([
Crop(H),
tv.transforms.ToTensor(),
tv.transforms.Normalize(mean=[0.5], std=[1.0])
])
(training_set_ok, ), (training_set_ko, ) = preproc.build_datasets(
path_training_ok, path_training_ko, dataset_name, transform, split=True)
training_loader_ok = DataLoader(training_set_ok,
batch_size=int(opt['batch_size'] / 2),
shuffle=True)
training_loader_ko = DataLoader(training_set_ko,
batch_size=int(opt['batch_size'] / 2),
shuffle=True)
validation_set, = preproc.build_datasets(path_validation_ok,
path_validation_ko, dataset_name,
transform)
val, test = preproc.split(validation_set, 0.7, random_state=56)
val_loader = DataLoader(val, batch_size=len(val), shuffle=False)
test_loader = DataLoader(test, batch_size=len(test), shuffle=False)