语义分割系列24-PointRend（pytorch实现）

PointRend: Image Segmentation as Rendering
论文链接：PointRend

本文将介绍：

• PointRend的原理
• PointRend代码实现
• PointRend在Camvid数据集上进行复现

---

---

引文

PointRend是何凯明团队发在2020年CVPR上的作品，何凯明出品，必属精品。

看到这篇论文的第一眼，还是很惊艳的。再读下去，发现还十分有趣。

PointRend的思想其实并不复杂，但是却很新颖，原因是将渲染领域的操作引入到分割领域（不管是实例分割还是语义分割）。PointRend本身可以理解为一个新颖的上采样操作。

在图像分割问题中，边缘的恢复\精确分割是比较麻烦的问题。在比较经典的一些语义分割模型中，无论是FCN、DeepLab还是PSPNet，在模型后都是接一个8倍或4倍的上采样操作来恢复图像，这对于物体边缘的预测自然是不利的，上采样会损失一些边缘信息。

因此呢，PointRend其实就是为上采样过程中精确恢复物体边缘的任务而生。

PointRend 思路

整体思路和PointRend实现步骤

首先先看整体的结构图：

1. 从PointRend的应用思路中可以看到，这里包含了两个阶段的特征处理，分别是fine-grained features和coarse prediction部分，如果主干网络是ResNet，那么fine-grained features就是ResNet的stage2输出，也就是4倍下采样时的精细分割结果，而coarse prediction就是检测头的预测结果（还未上采样还原成原图的结果）。
2. 从coarse prediction中挑选N个“难点”，也就是结果很有可能和周围点不一样的点（比如物体边缘的点）。
   对于每一个难点，获取他的“特征向量”，对于点特征向量（point features），主要由两部分组成，分别是fine-grained features的对应点和coarse prediction的对应点的特征向量，将这个两个特征向量拼接成一个向量。
3. 接着，通过一个MLP网络对这个“特征向量”进行预测，更新coarse prediction。也就相当于对这个难点进行新的预测，对他进行分类。

看完这个，我们就可以这么理解，将预测难的点（边缘点）提取出来，再提取其特征向量，经过MLP网络，将这个点的归属进行分类，然后提升这些点的分类准确率。这就是PointRend的思想。

我们再看细节部分的实现图：

对于一个coarse prediction(4x4大小)，将其上采样两倍(8x8大小，这里可以理解为检测头的输出)后，取了一些难分割的点（大多是边缘部分），取这些点的特征向量输入到MLP网络中，进行point prediction，得到每一个点的新类别，最后结果输出(8x8大小，边缘更加精确的结果)。

PointRend点选择策略

再了解了PointRend实现目标和总体思路后，我们还需对其细节做进一步了解。

既然PointRend是基于点的预测，那么这些点该如何采样？这将是一个问题，如果对全局点进行采样，那么计算量就过于大了。如果只想对预测困难点（物体边界）进行采样，那么这个采样该如何实现？带着这个问题，我们往下看。

对于点采样过程，需要对模型的Train过程和Inference过程做区分。

对于Inference过程

在Inference过程中，每个区域都通过迭代coarse-to-fine的方式来渲染。在每一次迭代过程中，PointRend都使用双线性差值将上一次的segmentation result进行上采样，然后在这个结果中选择N个不确定的点（分类概率接近0.5的点，也就是模型认为模棱两可的点）。这样就等于有目的的选到了N个分割困难的点，然后提取特征向量，经过MLP进行分类，得到新的segmentation result。

然后再上采样->提取点->MLP point prediction->…， 重复这个步骤，直到完成预测。这就是迭代coarse-to-fine的操作，也就是上图中figure 4的内容。

对于Train过程

对于Train过程的点采样操作，同样可以遵循Inference中的操作。但是作者发现，这样子采样对于梯度的传播不太友好，于是只能被迫选择其他的点采样策略——干脆就用随机采样的方式来进行采样。

1. 首先依据均匀分布随机取kN个点(k>1)
2. 然后上采样后，预测估计这些点的结果，再从kN个点中选取βN个点(0<β<1)

作者还选了几种采样方式，具体效果看下图。

PointRend效果

不管如何，PointRend对于物体的边缘恢复效果是很不错的，而且很灵活，可以作为上采样操作放置在很多网络之中，用来涨点。

PointRend的一些代码和实现

下面提供一些PointRend的代码（来自官方），以及在Camvid数据集上的测试代码。

点采样策略代码

import torch
import torch.nn.functional as F
def point_sample(input, point_coords, **kwargs):
    add_dim = False
    if point_coords.dim() == 3:
        add_dim = True
        point_coords = point_coords.unsqueeze(2)
    output = F.grid_sample(input, 2.0 * point_coords - 1.0, **kwargs)
    if add_dim:
        output = output.squeeze(3)
    return output


@torch.no_grad()
def sampling_points(mask, N, k=3, beta=0.75, training=True):
    assert mask.dim() == 4, "Dim must be N(Batch)CHW"
    device = mask.device
    B, _, H, W = mask.shape
    mask, _ = mask.sort(1, descending=True)

    if not training:
        H_step, W_step = 1 / H, 1 / W
        N = min(H * W, N)
        uncertainty_map = -1 * (mask[:, 0] - mask[:, 1])
        _, idx = uncertainty_map.view(B, -1).topk(N, dim=1)

        points = torch.zeros(B, N, 2, dtype=torch.float, device=device)
        points[:, :, 0] = W_step / 2.0 + (idx  % W).to(torch.float) * W_step
        points[:, :, 1] = H_step / 2.0 + (idx // W).to(torch.float) * H_step
        return idx, points

    over_generation = torch.rand(B, k * N, 2, device=device)
    over_generation_map = point_sample(mask, over_generation, align_corners=False)

    uncertainty_map = -1 * (over_generation_map[:, 0] - over_generation_map[:, 1])
    _, idx = uncertainty_map.topk(int(beta * N), -1)

    shift = (k * N) * torch.arange(B, dtype=torch.long, device=device)

    idx += shift[:, None]

    importance = over_generation.view(-1, 2)[idx.view(-1), :].view(B, int(beta * N), 2)
    coverage = torch.rand(B, N - int(beta * N), 2, device=device)
    return torch.cat([importance, coverage], 1).to(device)

ResNet+DeepLabv3 + PointRend实现代码

from torchvision.models._utils import IntermediateLayerGetter
from torchvision.models.segmentation._utils import _SimpleSegmentationModel
from torchvision.models.segmentation.deeplabv3 import DeepLabHead
from torchvision.models import resnet50, resnet101
import torch.nn as nn
from torchvision.models.resnet import ResNet, Bottleneck

class ResNetXX3(ResNet):
    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
                 groups=1, width_per_group=64, replace_stride_with_dilation=None,
                 norm_layer=None):
        super().__init__(block, layers, num_classes, zero_init_residual,
                         groups, width_per_group, replace_stride_with_dilation,
                         norm_layer)
        self.conv1 = nn.Conv2d(3, 64, 3, 1, 1, bias=False)
        nn.init.kaiming_normal_(self.conv1.weight, mode='fan_out', nonlinearity='relu')

def resnet53(pretrained=False, progress=True, **kwargs):
    r"""ResNet-50 model from
    `"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
    """
    return ResNetXX3(Bottleneck, [3, 4, 6, 3], **kwargs)

def resnet103(pretrained=False, progress=True, **kwargs):
    return ResNetXX3(Bottleneck, [3, 4, 23, 3], **kwargs)

class SmallDeepLab(_SimpleSegmentationModel):
    def forward(self, input_):
        result = self.backbone(input_)
        result["coarse"] = self.classifier(result["out"])
        return result

def deeplabv3(pretrained=False, resnet="res101", head_in_ch=2048, num_classes=21):
    resnet = {
        "res53":  resnet53,
        "res103": resnet103,
        "res50":  resnet50,
        "res101": resnet101
    }[resnet]

    net = SmallDeepLab(
        backbone=IntermediateLayerGetter(
            resnet(pretrained=False, replace_stride_with_dilation=[False, True, True]),
            return_layers={'layer2': 'res2', 'layer4': 'out'}
        ),
        classifier=DeepLabHead(head_in_ch, num_classes)
    )
    return net


if __name__ == "__main__":
    import torch
    x = torch.randn(3, 3, 224, 224).cuda()
    net = deeplabv3(False,num_classes=33).cuda()
    result = net(x)
    for k, v in result.items():
        print(k, v.shape)

import torch
import torch.nn as nn
import torch.nn.functional as F

class PointHead(nn.Module):
    def __init__(self,num_classes, in_c=512, k=3, beta=0.75):
        super().__init__()
        self.mlp = nn.Conv1d(in_c+num_classes, num_classes, 1)
        self.k = k
        self.beta = beta

    def forward(self, x, res2, out):
        """
        1. Fine-grained features are interpolated from res2 for DeeplabV3
        2. During training we sample as many points as there are on a stride 16 feature map of the input
        3. To measure prediction uncertainty
           we use the same strategy during training and inference: the difference between the most
           confident and second most confident class probabilities.
        """
        if not self.training:
            return self.inference(x, res2, out)

        points = sampling_points(out, x.shape[-1] // 16, self.k, self.beta)

        coarse = point_sample(out, points, align_corners=False)
        fine = point_sample(res2, points, align_corners=False)

        feature_representation = torch.cat([coarse, fine], dim=1)

        rend = self.mlp(feature_representation)

        return {"rend": rend, "points": points}

    @torch.no_grad()
    def inference(self, x, res2, out):
        """
        During inference, subdivision uses N=8096
        (i.e., the number of points in the stride 16 map of a 1024×2048 image)
        """
        num_points = 8096

        while out.shape[-1] != x.shape[-1]:
            out = F.interpolate(out, scale_factor=2, mode="bilinear", align_corners=True)

            points_idx, points = sampling_points(out, num_points, training=self.training)

            coarse = point_sample(out, points, align_corners=False)
            fine = point_sample(res2, points, align_corners=False)

            feature_representation = torch.cat([coarse, fine], dim=1)

            rend = self.mlp(feature_representation)

            B, C, H, W = out.shape
            points_idx = points_idx.unsqueeze(1).expand(-1, C, -1)
            out = (out.reshape(B, C, -1).scatter_(2, points_idx, rend).view(B, C, H, W))

        return {"fine": out}


class PointRend(nn.Module):
    def __init__(self, backbone, head):
        super().__init__()
        self.backbone = backbone
        self.head = head

    def forward(self, x):
        result = self.backbone(x)
        result.update(self.head(x, result["res2"], result["coarse"]))

        return result

if __name__ == "__main__":
    x = torch.randn(3, 3, 224, 224)
    net = PointRend(deeplabv3(False,num_classes=33), PointHead(num_classes=33))
    out = net(x)
    for k, v in out.items():
        print(k, v.shape)

在Camvid数据集上测试

# 导入库
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
import warnings
warnings.filterwarnings("ignore")
from PIL import Image
import numpy as np
import albumentations as A
from albumentations.pytorch.transforms import ToTensorV2
 
torch.manual_seed(17)
# 自定义数据集CamVidDataset
class CamVidDataset(torch.utils.data.Dataset):
    """CamVid Dataset. Read images, apply augmentation and preprocessing transformations.
    
    Args:
        images_dir (str): path to images folder
        masks_dir (str): path to segmentation masks folder
        class_values (list): values of classes to extract from segmentation mask
        augmentation (albumentations.Compose): data transfromation pipeline 
            (e.g. flip, scale, etc.)
        preprocessing (albumentations.Compose): data preprocessing 
            (e.g. noralization, shape manipulation, etc.)
    """
    
    def __init__(self, images_dir, masks_dir):
        self.transform = A.Compose([
            A.Resize(224, 224),
            A.HorizontalFlip(),
            A.VerticalFlip(),
            A.Normalize(),
            ToTensorV2(),
        ]) 
        self.ids = os.listdir(images_dir)
        self.images_fps = [os.path.join(images_dir, image_id) for image_id in self.ids]
        self.masks_fps = [os.path.join(masks_dir, image_id) for image_id in self.ids]
 
    
    def __getitem__(self, i):
        # read data
        image = np.array(Image.open(self.images_fps[i]).convert('RGB'))
        mask = np.array( Image.open(self.masks_fps[i]).convert('RGB'))
        image = self.transform(image=image,mask=mask)
        
        return image['image'], image['mask'][:,:,0]
        
    def __len__(self):
        return len(self.ids)
    
    
# 设置数据集路径
DATA_DIR = r'database/camvid/camvid/' # 根据自己的路径来设置
x_train_dir = os.path.join(DATA_DIR, 'train_images')
y_train_dir = os.path.join(DATA_DIR, 'train_labels')
x_valid_dir = os.path.join(DATA_DIR, 'valid_images')
y_valid_dir = os.path.join(DATA_DIR, 'valid_labels')
    
train_dataset = CamVidDataset(
    x_train_dir, 
    y_train_dir, 
)
val_dataset = CamVidDataset(
    x_valid_dir, 
    y_valid_dir, 
)
 
train_loader = DataLoader(train_dataset, batch_size=8, shuffle=True,drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=8, shuffle=True,drop_last=True)

model = PointRend(deeplabv3(False, num_classes=33), PointHead(num_classes=33)).cuda()

from d2l import torch as d2l
from tqdm import tqdm
import pandas as pd
import monai
# training loop 100 epochs
epochs_num = 100
# 选用SGD优化器来训练
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)
schedule = monai.optimizers.LinearLR(optimizer, end_lr=0.05, num_iter=int(epochs_num*0.75))
# 损失函数选用多分类交叉熵损失函数
lossf = nn.CrossEntropyLoss(ignore_index=255)


def evaluate_accuracy_gpu(net, data_iter, device=None):
    if isinstance(net, nn.Module):
        net.eval()  # Set the model to evaluation mode
        if not device:
            device = next(iter(net.parameters())).device
    # No. of correct predictions, no. of predictions
    metric = d2l.Accumulator(2)

    with torch.no_grad():
        for X, y in data_iter:
            if isinstance(X, list):
                # Required for BERT Fine-tuning (to be covered later)
                X = [x.to(device) for x in X]
            else:
                X = X.to(device)
            y = y.to(device)
            output = net(X)
            pred = F.interpolate(output["coarse"], X.shape[-2:], mode="bilinear", align_corners=True)
            metric.add(d2l.accuracy(pred, y), d2l.size(y))
    return metric[0] / metric[1]




# 训练函数
def train_ch13(net, train_iter, test_iter, loss, optimizer, num_epochs, schedule, swa_start=swa_start, devices=d2l.try_all_gpus()):
    timer, num_batches = d2l.Timer(), len(train_iter)
    animator = d2l.Animator(xlabel='epoch', xlim=[1, num_epochs], ylim=[0, 1], legend=['train loss', 'train acc', 'test acc'])
    net = nn.DataParallel(net, device_ids=devices).to(devices[0])
    # 用来保存一些训练参数

    loss_list = []
    train_acc_list = []
    test_acc_list = []
    epochs_list = []
    time_list = []
    lr_list = []
    

    for epoch in range(num_epochs):
        # Sum of training loss, sum of training accuracy, no. of examples,
        # no. of predictions
        metric = d2l.Accumulator(4)
        for i, (X, labels) in enumerate(train_iter):
            timer.start()

            if isinstance(X, list):
                X = [x.to(devices[0]) for x in X]
            else:
                X = X.to(devices[0])
            # y = labels.long().to(devices[0])
            gt = labels.squeeze_(1).to(devices[0], dtype=torch.long, non_blocking=True)

            net.train()
            optimizer.zero_grad()
            result = net(X)
            pred = F.interpolate(result["coarse"], X.shape[-2:], mode="bilinear", align_corners=True)

            seg_loss = F.cross_entropy(pred, gt, ignore_index=255)
            gt_points = point_sample(
                gt.float().unsqueeze(1),
                result["points"],
                mode="nearest",
                align_corners=False
            ).squeeze_(1).long()
            points_loss = F.cross_entropy(result["rend"], gt_points, ignore_index=255)  

            loss_sum = seg_loss + points_loss
            l = loss_sum
            loss_sum.sum().backward()
            optimizer.step()

            acc = d2l.accuracy(pred, gt)
            metric.add(l, acc, labels.shape[0], labels.numel())

            timer.stop()
            if (i + 1) % (num_batches // 5) == 0 or i == num_batches - 1:
                animator.add(epoch + (i + 1) / num_batches,(metric[0] / metric[2], metric[1] / metric[3], None))
                
        if optimizer.state_dict()['param_groups'][0]['lr']>0.05:
            schedule.step()

        test_acc = evaluate_accuracy_gpu(net, test_iter)

        animator.add(epoch + 1, (None, None, test_acc))

        print(f"epoch {epoch+1}/{epochs_num} --- loss {metric[0] / metric[2]:.3f} --- train acc {metric[1] / metric[3]:.3f} --- test acc {test_acc:.3f} --- lr {optimizer.state_dict()['param_groups'][0]['lr']} --- cost time {timer.sum()}")
        
        #---------保存训练数据---------------
        df = pd.DataFrame()
        loss_list.append(metric[0] / metric[2])
        train_acc_list.append(metric[1] / metric[3])
        test_acc_list.append(test_acc)
        epochs_list.append(epoch+1)
        time_list.append(timer.sum())
        lr_list.append(optimizer.state_dict()['param_groups'][0]['lr'])
        
        df['epoch'] = epochs_list
        df['loss'] = loss_list
        df['train_acc'] = train_acc_list
        df['test_acc'] = test_acc_list
        df["lr"] = lr_list
        df['time'] = time_list
        
        df.to_excel("savefile/PointRend_camvid.xlsx")
        #----------------保存模型------------------- 
        if np.mod(epoch+1, 5) == 0:
            torch.save(net.state_dict(), f'checkpoints/PointRend_{epoch+1}.pth')

    # 保存下最后的model
    torch.save(net.state_dict(), f'checkpoints/PointRend_last.pth')


train_ch13(model, train_loader, val_loader, lossf, optimizer, epochs_num, schedule=schedule)

结果