ckscr = cohen_kappa_score(y_true, y_pred)
f1scr = f1_score(y_true, y_pred, average="macro")
print("Cohen Kappa Score:", ckscr)
print("F1 Score: ", f1scr)
return (ckscr + f1scr) / 2
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
alphabet = preprocess.labels()
# Training data
X_train, X_te, y_train, y_te, train_ws = preprocess.load_data()
X_train = np.vstack((X_train, X_te))
y_train = np.hstack((y_train, y_te))
train_data = MyDataset(X_train, y_train,
preprocess.make_transform(mode="eval"))
train_loader = DataLoader(train_data,
batch_size=512,
shuffle=False,
num_workers=8,
pin_memory=True)
# Testing data
X_test, y_test, test_ws = preprocess.load_test_data()
test_data = MyDataset(X_test, y_test, preprocess.make_transform(mode="eval"))
test_loader = DataLoader(test_data,
batch_size=512,
shuffle=False,
num_workers=8,
pin_memory=True)
from modelling import MyDataset, get_model
from torch import nn
from torch import optim
from torch.utils.data import DataLoader
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Optimizer learning rate
learning_rate = 1e-2
# Epochs
epochs = 100
# Validation loss early stopping patience
# Number of epochs
patience = 10
X_train, X_val, y_train, y_val, ws = preprocess.load_data()
train_data = MyDataset(X_train, y_train, preprocess.make_transform("train"))
train_loader = DataLoader(
train_data, batch_size=128, shuffle=True, num_workers=8, pin_memory=torch.cuda.is_available()
)
test_data = MyDataset(X_val, y_val, preprocess.make_transform("eval"))
test_loader = DataLoader(
test_data, batch_size=512, shuffle=False, num_workers=8, pin_memory=torch.cuda.is_available()
)
base_dir = os.getcwd()
model_dir = os.path.join(base_dir, "Code", "model")
model_name = "ResNet"
model_path = os.path.join(model_dir, "sign_model.pth")
with open(os.path.join(model_dir, "model_specification"), "w") as ms:
ms.write(model_name)
my_classifier = get_model(model_name)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
labels = preprocess.labels()
# Load model
base_dir = os.getcwd()
model_dir = os.path.join(base_dir, "Code", "best_models", "model_09")
model_path = os.path.join(model_dir, "sign_model.pth")
with open(os.path.join(model_dir, "model_specification"), "r") as ms:
model_name = ms.readline().strip()
my_classifier = get_model(model_name)
my_classifier.load_state_dict(torch.load(model_path))
my_classifier.to(device)
my_classifier.eval()
transform = preprocess.make_transform(mode="predict")
plt.ion()
camera_url = "http://*****:*****@ip:8080/photo.jpg"
while True:
r = requests.get(camera_url)
f = io.BytesIO(r.content)
im = imread(f)
im = rotate(im, -90, resize=True)
im = im[420:-420, :]
im = transform(np.float32(im))
im = im.view(1, 3, 224, 224)
show_im = np.clip(im[0].numpy().transpose(1, 2, 0), 0, 1)
plt.clf()
plt.imshow(show_im)