Пример #1
0
    the_image = ImDict[image_name]
    image_size = np.shape(the_image)
    NRows = image_size[0]
    NCols = image_size[1]
    NBands = image_size[2]
    print("Lokalizacja obrazu: \t", filename)
    print("Nazwa obrazu:  \t\t\t", image_name)
    print("Rozmiar: \t\t\t\t", "wiersze: ", NRows, " kolumny: ", NCols,
          " zakresy: ", NBands)
    print("Ilośc pikseli (ilość kolumn * ilość wierszy): ", NRows * NCols)

    print()
    print("***   Converting image to uint8   ***")
    print("---------------------------------")
    # converted_image = mo.numpy_to_uint8(the_image)
    the_image = mo.numpy_to_uint8(the_image)

    print()
    print("***   Loading labels   ***")
    print("---------------------------------")
    # To juz jest w uint8
    filename_labels = 'C:/TestingCatalog/AI_data/Indian Pines/Indian_pines_gt.mat'
    ImDict_labels = io.loadmat(filename_labels)
    image_name_labels = 'indian_pines_gt'
    the_image_labels = ImDict_labels[image_name_labels]
    image_size_labels = np.shape(the_image_labels)
    NRows_labels = image_size_labels[0]
    NCols_labels = image_size_labels[1]
    '''
    import matplotlib.pyplot as plt
    plt.imshow(the_image_labels)
Пример #2
0
if __name__ == '__main__':
    print("START")
    filename = 'C:/TestingCatalog/AI_data/Indian Pines/Indian_pines_corrected.mat'
    ImDict = io.loadmat(filename)
    the_image = ImDict['indian_pines_corrected']

    image_size = np.shape(the_image)
    NRows = image_size[0]
    NCols = image_size[1]
    NBands = image_size[2]

    # przykład
    # img = open_image('92AV3C.lan')

    print(type(the_image))
    converted_image = mo.numpy_to_uint8(the_image)

    # Wyświetlanie obrazka
    view = imshow(converted_image, (29, 20, 11))
    view = imshow(the_image, (29, 20, 11))
    print(view)

    # Wyświtlanie kostki - nie działa
    # import spectral
    # spectral.settings.WX_GL_DEPTH_SIZE = 16
    # print(np.size(the_image))
    # print(np.shape(the_image))
    # smaller_img = the_image[0:30, 0:30, 0:30]
    # print(np.size(smaller_img))
    # print(np.shape(smaller_img))
    # view_cube(smaller_img, bands=[29, 20, 11])
def run_machine(input_net, first=True):
    to_file = False

    # Przekierowanie wyjścia do pliku
    if to_file:
        import sys
        orig_stdout = sys.stdout
        output_file = open('output_file.txt', 'w')
        sys.stdout = output_file

    print("START")
    start_time = time.time()
    print("Start time:  ", time.ctime(start_time))

    print()
    print("***   Loading data   ***")
    print("---------------------------------")
    filename = 'Indian_pines_corrected.mat'
    ImDict = io.loadmat(data_dir + filename)
    image_name = 'indian_pines_corrected'
    the_image = ImDict[image_name]
    image_size = np.shape(the_image)
    NRows = image_size[0]
    NCols = image_size[1]
    NBands = image_size[2]
    print("Lokalizacja obrazu: \t", data_dir + filename)
    print("Nazwa obrazu:  \t\t\t", image_name)
    print("Rozmiar: \t\t\t\t", "wiersze: ", NRows, " kolumny: ", NCols,
          " zakresy: ", NBands)
    print("Ilośc pikseli (ilość kolumn * ilość wierszy): ", NRows * NCols)

    print()
    print("***   Converting image to uint8   ***")
    print("---------------------------------")
    # converted_image = mo.numpy_to_uint8(the_image)
    the_image = mo.numpy_to_uint8(the_image)

    print()
    print("***   Loading labels   ***")
    print("---------------------------------")
    # To juz jest w uint8
    filename_labels = 'Indian_pines_gt.mat'
    ImDict_labels = io.loadmat(data_dir + filename_labels)
    image_name_labels = 'indian_pines_gt'
    the_image_labels = ImDict_labels[image_name_labels]
    image_size_labels = np.shape(the_image_labels)
    NRows_labels = image_size_labels[0]
    NCols_labels = image_size_labels[1]
    '''
    import matplotlib.pyplot as plt
    plt.imshow(the_image_labels)
    plt.show()
    '''
    labels = set()
    for row in the_image_labels:
        for element in row:
            labels.add(element)
    num_labels = len(labels)
    print("Lokalizacja obrazu: \t", filename_labels)
    print("Nazwa obrazu:  \t\t\t", image_name_labels)
    print("Rozmiar: \t\t\t\t", "wiersze: ", NRows_labels, " kolumny: ",
          NCols_labels)
    print("Ilośc etykiet: ", num_labels, " Etykiety: ", labels)

    print()
    print("***   Creating dataset and dataloader   ***")
    print("---------------------------------")
    import torch.utils.data as utils
    list_of_tensors = []
    for row in the_image:
        for element in row:
            list_of_tensors.append(torch.Tensor(element))

    list_of_tensors_labels = []
    for row in the_image_labels:
        for element in row:
            list_of_tensors_labels.append(torch.Tensor([element]))

    my_tensor = torch.stack(list_of_tensors)
    my_tensor_labels = torch.stack(list_of_tensors_labels)
    my_dataset = utils.TensorDataset(my_tensor, my_tensor_labels)
    my_dataloader = utils.DataLoader(my_dataset)
    print("Number of elements in dataset: ", my_dataset.__len__())

    print()
    print("***   Creating autoencoder   ***")
    print("---------------------------------")
    my_net = input_net
    print(my_net)
    params = list(my_net.parameters())
    print("Params size:  ", params.__len__())
    for parameter in params:
        print(len(parameter))

    if my_net.getName() != 'none':
        print()
        print("***   Creating optimizer   ***")
        print("---------------------------------")
        optimizer = torch.optim.Adam(my_net.parameters(), weight_decay=1e-5)
        criterion = nn.MSELoss()

    if first and my_net.getName() != 'none':
        print()
        print("***   Learning   ***")
        print("---------------------------------")
        from torch.autograd import Variable
        num_epochs = 3
        batch_size = 128
        learning_rate = 1e-3
        for epoch in range(num_epochs):
            for i, data in enumerate(my_dataloader):
                img, _ = data
                img = Variable(img).cpu()
                # ===================forward=====================
                output = my_net(img)
                loss = criterion(output, img)
                # ===================backward====================
                optimizer.zero_grad()
                loss.backward()
                optimizer.step()
            # ===================log========================
            print(type(loss))
            #print(len(loss))
            print(type(loss.item()))
            print('epoch [', epoch + 1, '/', num_epochs, '], loss:',
                  loss.item())
            # if epoch % 10 == 0:
            #    pic = to_img(output.cpu().data)

        print()
        print("***   Saving model to file   ***")
        print("---------------------------------")
        autoencoder_learned_file = './autoencoder_' + my_net.getType(
        ) + '_' + my_net.getName() + '.pth'
        torch.save(my_net.state_dict(), autoencoder_learned_file)

    if not first and my_net.getName() != 'none':
        print()
        print("***   Loading model from file   ***")
        print("---------------------------------")
        autoencoder_learned_file = './autoencoder_' + my_net.getType(
        ) + ' ' + my_net.getName() + '.pth'
        my_net.load_state_dict(torch.load(autoencoder_learned_file))
        my_net.eval()

    print()
    print("***   Checking code for one element   ***")
    print("---------------------------------")
    print(my_net.getCode(list_of_tensors[0]))

    print()
    print("***   Clustering   ***")
    print("---------------------------------")
    print()

    img_name = 'clustering_kmeans_' + my_net.getType(
    ) + '_autoencoder_' + my_net.getName() + '.png'
    clustering_kmeans(the_image, my_net, np.shape(the_image_labels), img_dir,
                      img_name)

    print("\nEND")
    end_time = time.time()
    print("End time:  ", time.ctime(end_time))
    print("Duration:  ", int(end_time - start_time), " seconds")

    # Closing file
    if to_file:
        sys.stdout = orig_stdout
        output_file.close()