View Source Bumblebee.Vision.ConvNextFeaturizer (Bumblebee v0.6.0)

ConvNeXT featurizer for image data.

Configuration

• :resize - whether to resize (and optionally center crop) the input to the given :size. Defaults to true

• :size - the size to resize the input to. If 384 or larger, the image is resized to (:size, :size). Otherwise, the shorter edge of the image is matched to :size / :crop_percentage, then image is cropped to :size. Only has an effect if :resize is true . Defaults to 224

• :resize_method - the resizing method, either of :nearest, :bilinear, :bicubic, :lanczos3, :lanczos5. Defaults to :bicubic

• :crop_percentage - the percentage of the image to crop. Only has an effect if :resize is true and :size < 384. Defaults to 0.875

• :normalize - whether or not to normalize the input with mean and standard deviation. Defaults to true

• :image_mean - the sequence of mean values for each channel, to be used when normalizing images. Defaults to [0.485, 0.456, 0.406]

• :image_std - the sequence of standard deviations for each channel, to be used when normalizing images. Defaults to [0.229, 0.224, 0.225]