Shortcuts

Migration of Model Settings

We update model settings in MMEdit 1.x. Important modifications are as following.

  • Remove pretrained fields.

  • Add train_cfg and test_cfg fields in model settings.

  • Add data_preprocessor fields. Normalization and color space transforms operations are moved from datasets transforms pipelines to data_preprocessor. We will introduce data_preprocessor later.

Original New
model = dict(
    type='BasicRestorer',  # Name of the model
    generator=dict(  # Config of the generator
        type='EDSR',  # Type of the generator
        in_channels=3,  # Channel number of inputs
        out_channels=3,  # Channel number of outputs
        mid_channels=64,  # Channel number of intermediate features
        num_blocks=16,  # Block number in the trunk network
        upscale_factor=scale, # Upsampling factor
        res_scale=1,  # Used to scale the residual in residual block
        rgb_mean=(0.4488, 0.4371, 0.4040),  # Image mean in RGB orders
        rgb_std=(1.0, 1.0, 1.0)),  # Image std in RGB orders
    pretrained=None,
    pixel_loss=dict(type='L1Loss', loss_weight=1.0, reduction='mean'))  # Config for pixel loss model training and testing settings
model = dict(
    type='BaseEditModel',  # Name of the model
    generator=dict(  # Config of the generator
        type='EDSRNet',  # Type of the generator
        in_channels=3,  # Channel number of inputs
        out_channels=3,  # Channel number of outputs
        mid_channels=64,  # Channel number of intermediate features
        num_blocks=16,  # Block number in the trunk network
        upscale_factor=scale, # Upsampling factor
        res_scale=1,  # Used to scale the residual in residual block
        rgb_mean=(0.4488, 0.4371, 0.4040),  # Image mean in RGB orders
        rgb_std=(1.0, 1.0, 1.0)),  # Image std in RGB orders
    pixel_loss=dict(type='L1Loss', loss_weight=1.0, reduction='mean')  # Config for pixel loss
    train_cfg=dict(),  # Config of training model.
    test_cfg=dict(),  # Config of testing model.
    data_preprocessor=dict(  # The Config to build data preprocessor
        type='EditDataPreprocessor', mean=[0., 0., 0.], std=[255., 255.,
                                                             255.]))

We refactor models in MMEdit 1.x. Important modifications are as following.

  • The models in MMedit 1.x is refactored to five parts: base_models, data_preprocessors, editors, layers and losses.

  • Add data_preprocessor module in models. Normalization and color space transforms operations are moved from datasets transforms pipelines to data_preprocessor. The data out from the data pipeline is transformed by this module and then fed into the model.

More details of models are shown in model guides.

Read the Docs v: zyh/doc-notfound-extend
Versions
master
latest
stable
zyh-doc-notfound-extend
Downloads
pdf
html
epub
On Read the Docs
Project Home
Builds

Free document hosting provided by Read the Docs.