|
MSE-CNN Implementation 1
Code database with the implementation of MSE-CNN, from the paper 'DeepQTMT: A Deep Learning Approach for Fast QTMT-based CU Partition of Intra-mode VVC'
|
Namespaces | |
| namespace | demo |
Functions | |
| demo.setup_model () | |
| Initializes and load the parameters of the MSE-CNN. | |
| demo.int2label (split) | |
| Obtain the string that corresponds to an integer value of the split. | |
| demo.draw_partition (img, split, cu_pos, cu_size) | |
| Draw partition in image based in the split outputed by the model. | |
| demo.split_fm (cu, cu_pos, split) | |
| Splits feature maps in specific way. | |
| demo.partition_img (img, img_yuv) | |
| Partitions a full 128x128 CTU and draws the partition in the original image. | |
| demo.pipeline (img, text) | |
| Pipeline to implement the functionalities to demonstrate the potential of the MSE-CNN. | |
| demo.main () | |
Variables | |
| str | demo.PATH_TO_COEFFS = "../../../model_coefficients/best_coefficients" |
| str | demo.LOAD_IMAGE_ERROR = "load_image_error.png" |
| list | demo.EXAMPLE_IMGS = ["example_img_1.jpeg", "example_img_2.jpeg"] |
| tuple | demo.CTU_SIZE = (128, 128) |
| demo.FIRST_CU_POS = torch.tensor([0, 0]).reshape(shape=(-1, 2)) | |
| demo.FIRST_CU_SIZE = torch.tensor([64, 64]).reshape(shape=(-1, 2)) | |
| str | demo.DEV = "cuda" if torch.cuda.is_available() else "cpu" |
| int | demo.QP = 32 |
| demo.model = None | |
| tuple | demo.COLOR = (0, 247, 255) |
| int | demo.LINE_THICKNESS = 1 |
| str | demo.DEFAULT_TEXT_FOR_COORDS = "Insert CTU position in the image..." |