asd

encoder_model = Encoder()
decoder_model = CaptionModel(vocab_size).to(device)
decoder_model.load_state_dict(torch.load(args.checkpoint))
for image_name in os.listdir("evaluate/images"):
    image = load_image(image_name, size=224)
    # convert the image pixels to a numpy array
    image = transforms.ToTensor()(image)
    # reshape data for the model
    image = image.unsqueeze(0)
    # prepare the image for the VGG model
    image = normalize_batch(image)
    features = encoder_model(image)
    predicted_sentence = generate_desc(decoder_model, tokenizer, features, max_length)
    print(predicted_sentence)
    plt.axis('off')
    plt.title(predicted_sentence)
    plt.savefig(os.path.join("evaluate/results/",
                                image_name+'_result.jpg'))