Method __init__

src/depth_pro/network/encoder.py:20–138

Initialize DepthProEncoder. The framework 1. creates an image pyramid, 2. generates overlapping patches with a sliding window at each pyramid level, 3. creates batched encodings via vision transformer backbones, 4. produces multi-resolution en

(
        self,
        dims_encoder: Iterable[int],
        patch_encoder: nn.Module,
        image_encoder: nn.Module,
        hook_block_ids: Iterable[int],
        decoder_features: int,
    )

    """

    def __init__(
        self,
        dims_encoder: Iterable[int],
        patch_encoder: nn.Module,
        image_encoder: nn.Module,
        hook_block_ids: Iterable[int],
        decoder_features: int,
    ):
        """Initialize DepthProEncoder.

        The framework
            1. creates an image pyramid,
            2. generates overlapping patches with a sliding window at each pyramid level,
            3. creates batched encodings via vision transformer backbones,
            4. produces multi-resolution encodings.

        Args:
        ----
            img_size: Backbone image resolution.
            dims_encoder: Dimensions of the encoder at different layers.
            patch_encoder: Backbone used for patches.
            image_encoder: Backbone used for global image encoder.
            hook_block_ids: Hooks to obtain intermediate features for the patch encoder model.
            decoder_features: Number of feature output in the decoder.

        """
        super().__init__()

        self.dims_encoder = list(dims_encoder)
        self.patch_encoder = patch_encoder
        self.image_encoder = image_encoder
        self.hook_block_ids = list(hook_block_ids)

        patch_encoder_embed_dim = patch_encoder.embed_dim
        image_encoder_embed_dim = image_encoder.embed_dim

        self.out_size = int(
            patch_encoder.patch_embed.img_size[0] // patch_encoder.patch_embed.patch_size[0]
        )

        def _create_project_upsample_block(
            dim_in: int,
            dim_out: int,
            upsample_layers: int,
            dim_int: Optional[int] = None,
        ) -> nn.Module:
            if dim_int is None:
                dim_int = dim_out
            # Projection.
            blocks = [
                nn.Conv2d(
                    in_channels=dim_in,
                    out_channels=dim_int,
                    kernel_size=1,
                    stride=1,
                    padding=0,
                    bias=False,
                )