support LightCDNet

README.md

@@ -37,6 +37,7 @@ Supported change detection model:
 - [x] [Changer (TGRS'2023)](configs/changer)
 - [x] [HANet (JSTARS'2023)](configs/hanet)
 - [x] [TinyCDv2 (Under Review)](configs/tinycd_v2)
+- [x] [LightCDNet (GRSL'2023)](configs/lightcdnet)
 - [x] [BAN (arXiv'2023)](configs/ban)
 - [x] [TTP (arXiv'2023)](configs/ttp)
 - [ ] ...

configs/_base_/models/lightcdnet.py

@@ -0,0 +1,54 @@
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+data_preprocessor = dict(
+    type='DualInputSegDataPreProcessor',
+    mean=[123.675, 116.28, 103.53] * 2,
+    std=[58.395, 57.12, 57.375] * 2,
+    bgr_to_rgb=True,
+    size_divisor=32,
+    pad_val=0,
+    seg_pad_val=255,
+    test_cfg=dict(size_divisor=32))
+model = dict(
+    type='DIEncoderDecoder',
+    data_preprocessor=data_preprocessor,
+    pretrained=None,
+    backbone=dict(
+        type='LightCDNet',
+        stage_repeat_num=[4, 8, 4],
+        net_type="small"),
+    neck=dict(
+        type='TinyFPN',
+        exist_early_x=True,
+        early_x_for_fpn=True,
+        custom_block='conv',
+        in_channels=[24, 48, 96, 192],
+        out_channels=48,
+        num_outs=4),
+    decode_head=dict(
+        type='DS_FPNHead',
+        in_channels=[48, 48, 48, 48],
+        in_index=[0, 1, 2, 3],
+        channels=48,
+        dropout_ratio=0.,
+        num_classes=2,
+        norm_cfg=norm_cfg,
+        align_corners=False,
+        loss_decode=dict(
+            type='mmseg.CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    auxiliary_head=dict(
+        type='mmseg.FCNHead',
+        in_channels=24,
+        in_index=0,
+        channels=24,
+        num_convs=1,
+        concat_input=False,
+        dropout_ratio=0.,
+        num_classes=2,
+        norm_cfg=norm_cfg,
+        align_corners=False,
+        loss_decode=dict(
+            type='mmseg.CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)),
+    # model training and testing settings
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))

configs/lightcdnet/README.md

@@ -0,0 +1,45 @@
+# LightCDNet
+
+[LightCDNet: Lightweight Change Detection Network Based on VHR Images](https://ieeexplore.ieee.org/document/10214556)
+
+## Introduction
+
+[Official Repo](https://github.com/NightSongs/LightCDNet)
+
+[Code Snippet](https://github.com/likyoo/open-cd/blob/main/opencd/models/backbones/lightcdnet.py)
+
+## Abstract
+Lightweight change detection models are essential for industrial applications and edge devices. Reducing the model size while maintaining high accuracy is a key challenge in developing lightweight change detection models. However, many existing methods oversimplify the model architecture, leading to a loss of information and reduced performance. Therefore, developing a lightweight model that can effectively preserve the input information is a challenging problem. To address this challenge, we propose LightCDNet, a novel lightweight change detection model that effectively preserves the input information. LightCDNet consists of an early fusion backbone network and a pyramid decoder for end-to-end change detection. The core component of LightCDNet is the Deep Supervised Fusion Module (DSFM), which guides the early fusion of primary features to improve performance. We evaluated LightCDNet on the LEVIR-CD dataset and found that it achieved comparable or better performance than state-of-the-art models while being 10–117 times smaller in size.
+
+<!-- [IMAGE] -->
+
+<div align=center>
+<img src="https://github.com/likyoo/open-cd/assets/44317497/cec088ca-cb45-4d32-8ebb-c0fd3b8d1a4c" width="90%"/>
+</div>
+
+
+```bibtex
+@ARTICLE{10214556,
+  author={Xing, Yuanjun and Jiang, Jiawei and Xiang, Jun and Yan, Enping and Song, Yabin and Mo, Dengkui},
+  journal={IEEE Geoscience and Remote Sensing Letters}, 
+  title={LightCDNet: Lightweight Change Detection Network Based on VHR Images}, 
+  year={2023},
+  volume={20},
+  number={},
+  pages={1-5},
+  doi={10.1109/LGRS.2023.3304309}}
+```
+
+## Results and models
+
+### LEVIR-CD
+
+|      Method      | Crop Size | Lr schd | \#Param (M) | MACs (G) | Precision | Recall | F1-Score |  IoU  | config                                                       |
+| :--------------: | :-------: | :-----: | :---------: | :------: | :-------: | :----: | :------: | :---: | ------------------------------------------------------------ |
+| LightCDNet-small |  256x256  |  40000  |    0.35     |   1.65   |   91.36   | 89.81  |  90.57   | 82.77 | [config](https://github.com/likyoo/open-cd/blob/main/configs/lightcdnet/lightcdnet_s_256x256_40k_levircd.py) |
+| LightCDNet-base  |  256x256  |  40000  |    1.32     |   3.22   |   92.12   | 90.43  |  91.27   | 83.94 | [config](https://github.com/likyoo/open-cd/blob/main/configs/lightcdnet/lightcdnet_b_256x256_40k_levircd.py) |
+| LightCDNet-large |  256x256  |  40000  |    2.82     |   5.94   |   92.43   | 90.45  |  91.43   | 84.21 | [config](https://github.com/likyoo/open-cd/blob/main/configs/lightcdnet/lightcdnet_l_256x256_40k_levircd.py) |
+
+
+- All metrics are based on the category "change".
+- All scores are computed on the test set.

configs/lightcdnet/lightcdnet_b_256x256_40k_levircd.py

@@ -0,0 +1,5 @@
+_base_ = ['./lightcdnet_s_256x256_40k_levircd.py']
+
+model = dict(
+    backbone=dict(net_type="base"),
+    neck=dict(in_channels=[24, 116, 232, 464]))

configs/lightcdnet/lightcdnet_l_256x256_40k_levircd.py

@@ -0,0 +1,5 @@
+_base_ = ['./lightcdnet_s_256x256_40k_levircd.py']
+
+model = dict(
+    backbone=dict(net_type="large"),
+    neck=dict(in_channels=[24, 176, 352, 704]))

configs/lightcdnet/lightcdnet_s_256x256_40k_levircd.py

@@ -0,0 +1,19 @@
+_base_ = [
+    '../_base_/models/lightcdnet.py',
+    '../common/standard_256x256_40k_levircd.py']
+
+model = dict(
+    decode_head=dict(
+        sampler=dict(type='mmseg.OHEMPixelSampler', thresh=0.7, min_kept=100000)))
+
+# optimizer
+optimizer = dict(
+    type='AdamW',
+    lr=0.003,
+    betas=(0.9, 0.999),
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='OptimWrapper',
+    optimizer=optimizer)

opencd/models/backbones/__init__.py

@@ -9,8 +9,10 @@
 from .hanet import HAN
 from .vit_tuner import VisionTransformerTurner
 from .vit_sam import ViTSAM_Custom
+from .lightcdnet import LightCDNet
 
 __all__ = ['IA_ResNetV1c', 'IA_ResNeSt', 'FC_EF', 'FC_Siam_diff', 
            'FC_Siam_conc', 'SNUNet_ECAM', 'TinyCD', 'IFN',
            'TinyNet', 'IA_MixVisionTransformer', 'HAN',
-           'VisionTransformerTurner', 'ViTSAM_Custom']
+           'VisionTransformerTurner', 'ViTSAM_Custom',
+           'LightCDNet']

opencd/models/backbones/lightcdnet.py

@@ -0,0 +1,227 @@
+# Copyright (c) Open-CD. All rights reserved.
+import torch
+import torch.nn as nn
+import numpy as np
+from mmcv.ops import CrissCrossAttention
+
+from mmseg.models.utils import LayerNorm2d
+from opencd.registry import MODELS
+
+
+class CCA(nn.Module):
+    """Criss-Cross Attention for Semantic Segmentation.
+
+    This head is the implementation of `CCNet
+    <https://arxiv.org/abs/1811.11721>`_.
+
+    Args:
+        recurrence (int): Number of recurrence of Criss Cross Attention
+            module. Default: 2.
+    """
+
+    def __init__(self, channels, recurrence=2):
+        super(CCA, self).__init__()
+        self.recurrence = recurrence
+        self.cca = CrissCrossAttention(channels)
+
+    def forward(self, x):
+        for _ in range(self.recurrence):
+            x = self.cca(x)
+        return x
+
+
+def channel_shuffle(x, groups=2):
+    bat_size, channels, w, h = x.shape
+    group_c = channels // groups
+    x = x.view(bat_size, groups, group_c, w, h)
+    x = torch.transpose(x, 1, 2).contiguous()
+    x = x.view(bat_size, -1, w, h)
+    return x
+
+
+class ShuffleBlock(nn.Module):
+
+    def __init__(self, in_c, out_c, downsample=False):
+        super(ShuffleBlock, self).__init__()
+        self.downsample = downsample
+        half_c = out_c // 2
+        if downsample:
+            self.branch1 = nn.Sequential(
+                # 3*3 dw conv, stride = 2
+                nn.Conv2d(in_c, in_c, 3, 2, 1, groups=in_c, bias=False),
+                nn.BatchNorm2d(in_c),
+                # 1*1 pw conv
+                nn.Conv2d(in_c, half_c, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(half_c),
+                nn.ReLU(True))
+
+            self.branch2 = nn.Sequential(
+                # 1*1 pw conv
+                nn.Conv2d(in_c, half_c, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(half_c),
+                nn.ReLU(True),
+                # 3*3 dw conv, stride = 2
+                nn.Conv2d(half_c, half_c, 3, 2, 1, groups=half_c, bias=False),
+                nn.BatchNorm2d(half_c),
+                # 1*1 pw conv
+                nn.Conv2d(half_c, half_c, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(half_c),
+                nn.ReLU(True))
+
+        else:
+            assert in_c == out_c
+
+            self.branch2 = nn.Sequential(
+                # 1*1 pw conv
+                nn.Conv2d(half_c, half_c, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(half_c),
+                nn.ReLU(True),
+                # 3*3 dw conv, stride = 1
+                nn.Conv2d(half_c, half_c, 3, 1, 1, groups=half_c, bias=False),
+                nn.BatchNorm2d(half_c),
+                # 1*1 pw conv
+                nn.Conv2d(half_c, half_c, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(half_c),
+                nn.ReLU(True))
+
+    def forward(self, x):
+        out = None
+        if self.downsample:
+            # if it is downsampling, we don't need to do channel split
+            out = torch.cat((self.branch1(x), self.branch2(x)), 1)
+        else:
+            # channel split
+            channels = x.shape[1]
+            c = channels // 2
+            x1 = x[:, :c, :, :]
+            x2 = x[:, c:, :, :]
+            out = torch.cat((x1, self.branch2(x2)), 1)
+
+        return channel_shuffle(out, 2)
+
+
+class TimeAttention(nn.Module):
+
+    def __init__(self, channels):
+        super(TimeAttention, self).__init__()
+        self.avg_pool = nn.AdaptiveAvgPool2d(1)
+        attn_channels = channels // 16
+        attn_channels = max(attn_channels, 8)
+        self.mlp = nn.Sequential(
+            nn.Conv2d(channels * 2, attn_channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(attn_channels),
+            nn.ReLU(),
+            nn.Conv2d(attn_channels, channels * 2, kernel_size=1, bias=False),
+        )
+
+    def forward(self, x1, x2):
+        x = torch.cat((x1, x2), dim=1)
+        x = self.avg_pool(x)
+        y = self.mlp(x)
+        B, C, H, W = y.size()
+        x1_attn, x2_attn = y.reshape(B, 2, C // 2, H, W).transpose(0, 1)
+        x1_attn = torch.sigmoid(x1_attn)
+        x2_attn = torch.sigmoid(x2_attn)
+        x1 = x1 * x1_attn + x1
+        x2 = x2 * x2_attn + x2
+        return x1, x2
+
+
+class shuffle_fusion(nn.Module):
+
+    def __init__(self, channels, block_num=2):
+        super().__init__()
+
+        self.stages = []
+        self.stages.append(
+            nn.Sequential(
+                nn.Conv2d(channels, channels * 4, kernel_size=1, bias=False),
+                nn.BatchNorm2d(channels * 4), nn.ReLU()))
+        for i in range(block_num):
+            self.stages.append(
+                ShuffleBlock(channels * 4, channels * 4, downsample=False))
+
+        self.stages = nn.Sequential(*self.stages)
+
+        self.single_conv = nn.Sequential(
+            nn.Conv2d(channels * 4, channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(channels), nn.ReLU())
+
+        self.time_attn = TimeAttention(channels)
+
+        self.final_conv = nn.Sequential(
+            nn.Conv2d(channels * 2, channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(channels), nn.ReLU())
+
+    def forward_single(self, x):
+        identity = x
+        x = self.stages(x)
+        x = self.single_conv(x)
+        x = identity + x
+        return x
+
+    def forward(self, x1, x2):
+        x1 = self.forward_single(x1)
+        x2 = self.forward_single(x2)
+        x1, x2 = self.time_attn(x1, x2)
+        x = self.final_conv(channel_shuffle(torch.cat((x1, x2), dim=1)))
+        return x
+
+
+@MODELS.register_module()
+class LightCDNet(nn.Module):
+
+    def __init__(self, stage_repeat_num, net_type="small"):
+        super(LightCDNet, self).__init__()
+
+        index_list = stage_repeat_num.copy()
+        index_list[0] = index_list[0] - 1
+        self.index_list = list(np.cumsum(index_list))
+        if net_type == "small":
+            self.out_channels = [24, 48, 96, 192]
+            self.block_num = 4
+        elif net_type == "base":
+            self.out_channels = [24, 116, 232, 464]
+            self.block_num = 8
+        elif net_type == "large":
+            self.out_channels = [24, 176, 352, 704]
+            self.block_num = 16
+        else:
+            print("the model type is error!")
+
+        self.conv1 = nn.Sequential(
+            nn.Conv2d(3, self.out_channels[0], 3, 2, 1, bias=False),
+            LayerNorm2d(self.out_channels[0]), nn.GELU())
+
+        self.fusion_conv = shuffle_fusion(
+            self.out_channels[0], block_num=self.block_num)
+
+        in_c = self.out_channels[0]
+
+        self.stages = []
+        for stage_idx in range(len(stage_repeat_num)):
+            out_c = self.out_channels[1 + stage_idx]
+            repeat_num = stage_repeat_num[stage_idx]
+            for i in range(repeat_num):
+                if i == 0:
+                    self.stages.append(
+                        ShuffleBlock(in_c, out_c, downsample=True))
+                else:
+                    self.stages.append(
+                        ShuffleBlock(in_c, in_c, downsample=False))
+                in_c = out_c
+            self.stages.append(CCA(channels=out_c, recurrence=2))
+
+        self.stages = nn.Sequential(*self.stages)
+
+    def forward(self, x1, x2):
+        x1 = self.conv1(x1)
+        x2 = self.conv1(x2)
+        x = self.fusion_conv(x1, x2)
+        outs = [x]
+
+        for i in range(len(self.stages)):
+            x = self.stages[i](x)
+            if i in self.index_list:
+                outs.append(x)
+        return outs

opencd/models/decode_heads/__init__.py

@@ -8,7 +8,8 @@
 from .ban_head import BitemporalAdapterHead
 from .ban_utils import BAN_MLPDecoder, BAN_BITHead
 from .mlpseg_head import MLPSegHead
+from .ds_fpn_head import DS_FPNHead
 
 __all__ = ['BITHead', 'Changer', 'IdentityHead', 'DSIdentityHead', 'TinyHead',
            'STAHead', 'MultiHeadDecoder', 'GeneralSCDHead', 'BitemporalAdapterHead',
-           'BAN_MLPDecoder', 'BAN_BITHead', 'MLPSegHead']
+           'BAN_MLPDecoder', 'BAN_BITHead', 'MLPSegHead', 'DS_FPNHead']