models/networks/xception.py

""" 
Creates an Xception Model as defined in:

Francois Chollet
Xception: Deep Learning with Depthwise Separable Convolutions
https://arxiv.org/pdf/1610.02357.pdf

This weights ported from the Keras implementation. Achieves the following performance on the validation set:

Loss:0.9173 Prec@1:78.892 Prec@5:94.292

REMEMBER to set your image size to 3x299x299 for both test and validation

normalize = transforms.Normalize(mean=[0.5, 0.5, 0.5],
                                  std=[0.5, 0.5, 0.5])

The resize parameter of the validation transform should be 333, and make sure to center crop at 299x299
"""
import math

import torch.nn as nn
import torch.nn.functional as F
import torch.utils.model_zoo as model_zoo
from torch.nn import init
import torch

from ..builder import MODELS
from .common import conv_block, BN_MOMENTUM


model_urls = {
    'xception':'https://www.dropbox.com/s/1hplpzet9d7dv29/xception-c0a72b38.pth.tar?dl=1'
}


class SeparableConv2d(nn.Module):
    def __init__(self,in_channels,out_channels,kernel_size=1,stride=1,padding=0,dilation=1,bias=False):
        super(SeparableConv2d,self).__init__()

        self.conv1 = nn.Conv2d(in_channels,in_channels,kernel_size,stride,padding,dilation,groups=in_channels,bias=bias)
        self.pointwise = nn.Conv2d(in_channels,out_channels,1,1,0,1,1,bias=bias)
    
    def forward(self,x):
        x = self.conv1(x)
        x = self.pointwise(x)
        return x


class Block(nn.Module):
    def __init__(self,in_filters,out_filters,reps,strides=1,start_with_relu=True,grow_first=True):
        super(Block, self).__init__()

        if out_filters != in_filters or strides!=1:
            self.skip = nn.Conv2d(in_filters,out_filters,1,stride=strides, bias=False)
            self.skipbn = nn.BatchNorm2d(out_filters)
        else:
            self.skip=None
        
        self.relu = nn.ReLU(inplace=True)
        rep=[]

        filters=in_filters
        if grow_first:
            rep.append(self.relu)
            rep.append(SeparableConv2d(in_filters,out_filters,3,stride=1,padding=1,bias=False))
            rep.append(nn.BatchNorm2d(out_filters))
            filters = out_filters

        for i in range(reps-1):
            rep.append(self.relu)
            rep.append(SeparableConv2d(filters,filters,3,stride=1,padding=1,bias=False))
            rep.append(nn.BatchNorm2d(filters))
        
        if not grow_first:
            rep.append(self.relu)
            rep.append(SeparableConv2d(in_filters,out_filters,3,stride=1,padding=1,bias=False))
            rep.append(nn.BatchNorm2d(out_filters))

        if not start_with_relu:
            rep = rep[1:]
        else:
            rep[0] = nn.ReLU(inplace=False)

        if strides != 1:
            rep.append(nn.MaxPool2d(3,strides,1))
        self.rep = nn.Sequential(*rep)

    def forward(self,inp):
        x = self.rep(inp)

        if self.skip is not None:
            skip = self.skip(inp)
            skip = self.skipbn(skip)
        else:
            skip = inp

        x+=skip
        return x


@MODELS.register_module()
class Xception(nn.Module):
    """
    Xception optimized for the ImageNet dataset, as specified in
    https://arxiv.org/pdf/1610.02357.pdf
    """
    def __init__(self, 
                 heads, 
                 head_conv=64, 
                 cls_based_hm=True,
                 dropout_prob=0.5,
                 **kwargs):
        """ Constructor
        Args:
            num_classes: number of classes
        """
        self.heads = heads
        self.head_conv = head_conv
        self.cls_based_hm = cls_based_hm
        self.dropout_prob = dropout_prob
        super(Xception, self).__init__()
        
        self.conv1 = nn.Conv2d(3, 32, 3,2, 0, bias=False)
        self.bn1 = nn.BatchNorm2d(32)
        self.relu = nn.ReLU(inplace=True)

        self.conv2 = nn.Conv2d(32,64,3,bias=False)
        self.bn2 = nn.BatchNorm2d(64)
        #do relu here

        self.block1=Block(64,128,2,2,start_with_relu=False,grow_first=True)
        self.block2=Block(128,256,2,2,start_with_relu=True,grow_first=True)
        self.block3=Block(256,728,2,2,start_with_relu=True,grow_first=True)

        self.block4=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block5=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block6=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block7=Block(728,728,3,1,start_with_relu=True,grow_first=True)

        self.block8=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block9=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block10=Block(728,728,3,1,start_with_relu=True,grow_first=True)
        self.block11=Block(728,728,3,1,start_with_relu=True,grow_first=True)

        self.block12=Block(728,1024,2,2,start_with_relu=True,grow_first=False)

        self.conv3 = SeparableConv2d(1024,1536,3,1,1)
        self.bn3 = nn.BatchNorm2d(1536)

        #do relu here
        self.conv4 = SeparableConv2d(1536,2048,3,1,1)
        self.bn4 = nn.BatchNorm2d(2048)
        
        self.dropout = nn.Dropout2d(p=self.dropout_prob)
    
        self.conv_block_1 = conv_block(2048, 256, (3,3), padding=1)
        self.deconv_1 = nn.Sequential(
            nn.ConvTranspose2d(
                in_channels=256,
                out_channels=256,
                kernel_size=(4,4),
                stride=2,
                padding=1,
                output_padding=0,
                bias=False),
            nn.BatchNorm2d(256, momentum=BN_MOMENTUM),
            nn.ReLU(inplace=True)
        )

        self.conv_block_2 = conv_block(256, 256, (3,3), padding=1)
        self.deconv_2 = nn.Sequential(
            nn.ConvTranspose2d(
                in_channels=256,
                out_channels=128,
                kernel_size=(4,4),
                stride=2,
                padding=1,
                output_padding=0,
                bias=False),
            nn.BatchNorm2d(128, momentum=BN_MOMENTUM),
            nn.ReLU(inplace=True)
        )
        
        self.conv_block_3 = conv_block(128, 128, (3,3), padding=1)
        self.deconv_3 = nn.Sequential(
            nn.ConvTranspose2d(
                in_channels=128,
                out_channels=64,
                kernel_size=(4,4),
                stride=2,
                padding=1,
                output_padding=0,
                bias=False),
            nn.BatchNorm2d(64, momentum=BN_MOMENTUM),
            nn.ReLU(inplace=True)
        )

        for head in sorted(self.heads):
            num_output = self.heads[head]
            if self.head_conv > 0:
                if head != 'cls':
                    fc = nn.Sequential(
                        nn.Conv2d(64, self.head_conv,
                        kernel_size=3, padding=1, bias=False),
                        nn.BatchNorm2d(self.head_conv),
                        nn.ReLU(inplace=True),
                        nn.Conv2d(self.head_conv, num_output, 
                        kernel_size=1, stride=1, padding=0)
                    )
                else:
                    if self.cls_based_hm:
                        fc = nn.Sequential(
                                nn.AdaptiveAvgPool2d(head_conv//4),
                                nn.Flatten(),
                                nn.Linear((head_conv//4)**2, head_conv, bias=False),
                                nn.BatchNorm1d(head_conv, momentum=BN_MOMENTUM),
                                nn.ReLU(inplace=True),
                                nn.Linear(head_conv, num_output, bias=True),
                                nn.Sigmoid()
                            )
                    else:
                        fc = nn.Sequential(
                            nn.Conv2d(64, head_conv, kernel_size=3,
                                      padding=1, bias=False),
                            nn.BatchNorm2d(head_conv, momentum=BN_MOMENTUM),
                            nn.ReLU(inplace=True),
                            nn.Conv2d(head_conv, num_output, kernel_size=1,
                                      stride=1, padding=0, bias=False),
                            nn.BatchNorm2d(num_output),
                            # nn.ReLU(inplace=True),
                            nn.AdaptiveAvgPool2d(head_conv//4),
                            nn.Flatten(),
                            nn.Linear((head_conv//4)**2, head_conv, bias=False),
                            nn.BatchNorm1d(head_conv, momentum=BN_MOMENTUM),
                            nn.ReLU(inplace=True),
                            nn.Linear(head_conv, num_output, bias=True),
                            nn.Sigmoid()
                        )
            else:
                fc = nn.Conv2d(
                    in_channels=64,
                    out_channels=num_output,
                    kernel_size=1,
                    stride=1,
                    padding=0
                )
            self.__setattr__(head, fc)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        
        x = self.conv2(x)
        x = self.bn2(x)
        x = self.relu(x)
        
        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)
        x = self.block4(x)
        x = self.block5(x)
        x = self.block6(x)
        x = self.block7(x)
        x = self.block8(x)
        x = self.block9(x)
        x = self.block10(x)
        x = self.block11(x)
        x = self.block12(x)
        
        x = self.conv3(x)
        x = self.bn3(x)
        x = self.relu(x)
        
        x = self.conv4(x)
        x = self.bn4(x)
        x = self.relu(x)
        
        x = self.dropout(x)

        x = self.conv_block_1(x)
        x = self.deconv_1(x)

        x = self.conv_block_2(x)
        x = self.deconv_2(x)

        x = self.conv_block_3(x)
        x = self.deconv_3(x)
        
        ret = {}
        x1_hm = None
        for head in self.heads:
            if not self.cls_based_hm or head != 'cls':
                ret[head] = self.__getattr__(head)(x)
                if head == 'hm':
                    x1_hm = ret[head]
            else:
                assert 'hm' in ret.keys(), "Other heads need features from heatmap, please check it!"
                ret[head] = self.__getattr__(head)(x1_hm)
        return [ret]
    
    def init_weights(self, pretrained=False):
        if not pretrained:
            for m in self.modules():
                if isinstance(m, nn.Conv2d):
                    n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                    m.weight.data.normal_(0, math.sqrt(2. / n))
                elif isinstance(m, nn.BatchNorm2d):
                    m.weight.data.fill_(1)
                    m.bias.data.zero_()
                elif isinstance(m, nn.ConvTranspose2d):
                    n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                    m.weight.data.normal_(0, math.sqrt(2. / n))
                    if self.deconv_with_bias:
                        nn.init.constant_(m.bias, 0)
        else:
            self.load_state_dict(model_zoo.load_url(model_urls['xception']), strict=False)

        # Init head parameters
        for head in self.heads:
            final_layer = self.__getattr__(head)
            for i, m in enumerate(final_layer.modules()):
                prior = 1/71
        #         if isinstance(m, nn.Conv2d):
        #             if m.weight.shape[0] == self.heads[head]:
        #                 if 'hm' in head:
        #                     # nn.init.constant_(m.bias, -2.19)
        #                     nn.init.constant_(m.bias, -math.log((1-prior)/prior))
        #                 else:
        #                     nn.init.normal_(m.weight, std=0.001)
        #                     # nn.init.constant_(m.bias, 0)
                if isinstance(m, nn.Linear):
                    if m.weight.shape[0] == self.heads[head]:
                        nn.init.constant_(m.bias, -math.log((1-prior)/prior))
                    # else:
                        # n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                        # m.weight.data.normal_(0, math.sqrt(2. / n))
        #                 # nn.init.constant_(m.bias, 0)