models/utils/dense.py

# -*- coding: utf-8 -*-
# Dense Block as defined in:
# Huang, Gao, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q. Weinberger.
# "Densely connected convolutional networks." In Proceedings of the IEEE conference
# on computer vision and pattern recognition, pp. 4700-4708. 2017.
#
# Code Snippet adapted from HoverNet implementation (https://github.com/vqdang/hover_net)
#
# @ Fabian Hörst, [email protected]
# Institute for Artifical Intelligence in Medicine,
# University Medicine Essen


import torch
import torch.nn as nn

from collections import OrderedDict


class DenseBlock(nn.Module):
    """Dense Block as defined in:

    Huang, Gao, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q. Weinberger.
    "Densely connected convolutional networks." In Proceedings of the IEEE conference
    on computer vision and pattern recognition, pp. 4700-4708. 2017.

    Only performs `valid` convolution.

    """

    def __init__(self, in_ch, unit_ksize, unit_ch, unit_count, split=1):
        super(DenseBlock, self).__init__()
        assert len(unit_ksize) == len(unit_ch), "Unbalance Unit Info"

        self.nr_unit = unit_count
        self.in_ch = in_ch
        self.unit_ch = unit_ch

        # ! For inference only so init values for batchnorm may not match tensorflow
        unit_in_ch = in_ch
        self.units = nn.ModuleList()
        for idx in range(unit_count):
            self.units.append(
                nn.Sequential(
                    OrderedDict(
                        [
                            ("preact_bna/bn", nn.BatchNorm2d(unit_in_ch, eps=1e-5)),
                            ("preact_bna/relu", nn.ReLU(inplace=True)),
                            (
                                "conv1",
                                nn.Conv2d(
                                    unit_in_ch,
                                    unit_ch[0],
                                    unit_ksize[0],
                                    stride=1,
                                    padding=0,
                                    bias=False,
                                ),
                            ),
                            ("conv1/bn", nn.BatchNorm2d(unit_ch[0], eps=1e-5)),
                            ("conv1/relu", nn.ReLU(inplace=True)),
                            # ('conv2/pool', TFSamepaddingLayer(ksize=unit_ksize[1], stride=1)),
                            (
                                "conv2",
                                nn.Conv2d(
                                    unit_ch[0],
                                    unit_ch[1],
                                    unit_ksize[1],
                                    groups=split,
                                    stride=1,
                                    padding=0,
                                    bias=False,
                                ),
                            ),
                        ]
                    )
                )
            )
            unit_in_ch += unit_ch[1]

        self.blk_bna = nn.Sequential(
            OrderedDict(
                [
                    ("bn", nn.BatchNorm2d(unit_in_ch, eps=1e-5)),
                    ("relu", nn.ReLU(inplace=True)),
                ]
            )
        )

    def out_ch(self):
        return self.in_ch + self.nr_unit * self.unit_ch[-1]

    def init_weights(self):
        """Kaiming (HE) initialization for convolutional layers and constant initialization for normalization and linear layers"""
        for m in self.modules():
            classname = m.__class__.__name__

            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")

            if "norm" in classname.lower():
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

            if "linear" in classname.lower():
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)

    def forward(self, prev_feat):
        for idx in range(self.nr_unit):
            new_feat = self.units[idx](prev_feat)
            prev_feat = crop_to_shape(prev_feat, new_feat)
            prev_feat = torch.cat([prev_feat, new_feat], dim=1)
        prev_feat = self.blk_bna(prev_feat)

        return prev_feat


# helper functions for cropping
def crop_op(x, cropping, data_format="NCHW"):
    """Center crop image.

    Args:
        x: input image
        cropping: the substracted amount
        data_format: choose either `NCHW` or `NHWC`

    """
    crop_t = cropping[0] // 2
    crop_b = cropping[0] - crop_t
    crop_l = cropping[1] // 2
    crop_r = cropping[1] - crop_l
    if data_format == "NCHW":
        x = x[:, :, crop_t:-crop_b, crop_l:-crop_r]
    else:
        x = x[:, crop_t:-crop_b, crop_l:-crop_r, :]
    return x


def crop_to_shape(x, y, data_format="NCHW"):
    """Centre crop x so that x has shape of y. y dims must be smaller than x dims.

    Args:
        x: input array
        y: array with desired shape.

    """
    assert (
        y.shape[0] <= x.shape[0] and y.shape[1] <= x.shape[1]
    ), "Ensure that y dimensions are smaller than x dimensions!"

    x_shape = x.size()
    y_shape = y.size()
    if data_format == "NCHW":
        crop_shape = (x_shape[2] - y_shape[2], x_shape[3] - y_shape[3])
    else:
        crop_shape = (x_shape[1] - y_shape[1], x_shape[2] - y_shape[2])
    return crop_op(x, crop_shape, data_format)