山西省网站建设seo引擎优化公司

（a）Mask RCNN总体流程

一.Mask RCNN 架构

自己整理了一份Mask RCNN架构图如下，其中绿色模块只有推理过程才会涉及。

核心模块包括：数据预处理，骨干网络，区域提议网络，FastRCNN分支，Mask分支，数据后处理等。

二.网络核心流程

class FasterRCNNBase(nn.Module):def __init__(self, backbone, rpn, roi_heads, transform):super(FasterRCNNBase, self).__init__()self.transform = transformself.backbone = backboneself.rpn = rpnself.roi_heads = roi_heads# used only on torchscript modeself._has_warned = False@torch.jit.unuseddef eager_outputs(self, losses, detections):# type: (Dict[str, Tensor], List[Dict[str, Tensor]]) -> Union[Dict[str, Tensor], List[Dict[str, Tensor]]]if self.training:return lossesreturn detectionsdef forward(self, images, targets=None):# type: (List[Tensor], Optional[List[Dict[str, Tensor]]]) -> Tuple[Dict[str, Tensor], List[Dict[str, Tensor]]]if self.training and targets is None:raise ValueError("In training mode, targets should be passed")if self.training:assert targets is not Nonefor target in targets:         # 进一步判断传入的target的boxes参数是否符合规定boxes = target["boxes"]if isinstance(boxes, torch.Tensor):if len(boxes.shape) != 2 or boxes.shape[-1] != 4:raise ValueError("Expected target boxes to be a tensor""of shape [N, 4], got {:}.".format(boxes.shape))else:raise ValueError("Expected target boxes to be of type ""Tensor, got {:}.".format(type(boxes)))original_image_sizes = torch.jit.annotate(List[Tuple[int, int]], [])for img in images:val = img.shape[-2:]assert len(val) == 2  # 防止输入的是个一维向量original_image_sizes.append((val[0], val[1]))# original_image_sizes = [img.shape[-2:] for img in images]images, targets = self.transform(images, targets)  # 对图像进行预处理# print(images.tensors.shape)features = self.backbone(images.tensors)  # 将图像输入backbone得到特征图if isinstance(features, torch.Tensor):  # 若只在一层特征层上预测，将feature放入有序字典中，并编号为‘0’features = OrderedDict([('0', features)])  # 若在多层特征层上预测，传入的就是一个有序字典# 将特征层以及标注target信息传入rpn中# proposals: List[Tensor], Tensor_shape: [num_proposals, 4],# 每个proposals是绝对坐标，且为(x1, y1, x2, y2)格式proposals, proposal_losses = self.rpn(images, features, targets)# 将rpn生成的数据以及标注target信息传入fast rcnn后半部分detections, detector_losses = self.roi_heads(features, proposals, images.image_sizes, targets)# 对网络的预测结果进行后处理（主要将bboxes还原到原图像尺度上）detections = self.transform.postprocess(detections, images.image_sizes, original_image_sizes)losses = {}losses.update(detector_losses)losses.update(proposal_losses)if torch.jit.is_scripting():if not self._has_warned:warnings.warn("RCNN always returns a (Losses, Detections) tuple in scripting")self._has_warned = Truereturn losses, detectionselse:return self.eager_outputs(losses, detections)# if self.training:#     return losses## return detections

FasterRCNNBase是RCNN检测算法的基类，FasterRCNN类要继承FasterRCNNBase类，而MaskRCNN类又要继承FasterRCNN类，所以当实例化一个model并传入数据x时，会调用FasterRCNNBase的forward函数：

model = MaskRCNN(backbone,num_classes)
model(images,targets)

FasterRCNNBase的 init() 函数:

    def __init__(self, backbone, rpn, roi_heads, transform):super(FasterRCNNBase, self).__init__()self.transform = transformself.backbone = backboneself.rpn = rpnself.roi_heads = roi_heads# used only on torchscript modeself._has_warned = False

传入参数包括：
（1）backbone：
resnet50
resnet101
resnet50+fpn
resnet101+fpn
（2）rpn:
区域提议网络
（3）roi_haeds:
box roi pooling/align
two MLP head
box predictor
mask roi pool
mask head
mask predictor
（4）transforms：
GeneraRCNNtransforms类的实例，用于数据预处理

FasterRCNNBase的 forward() 函数:

    def forward(self, images, targets=None):# type: (List[Tensor], Optional[List[Dict[str, Tensor]]]) -> Tuple[Dict[str, Tensor], List[Dict[str, Tensor]]]if self.training and targets is None:raise ValueError("In training mode, targets should be passed")if self.training:assert targets is not Nonefor target in targets:         # 进一步判断传入的target的boxes参数是否符合规定boxes = target["boxes"]if isinstance(boxes, torch.Tensor):if len(boxes.shape) != 2 or boxes.shape[-1] != 4:raise ValueError("Expected target boxes to be a tensor""of shape [N, 4], got {:}.".format(boxes.shape))else:raise ValueError("Expected target boxes to be of type ""Tensor, got {:}.".format(type(boxes)))original_image_sizes = torch.jit.annotate(List[Tuple[int, int]], [])for img in images:val = img.shape[-2:]assert len(val) == 2  # 防止输入的是个一维向量original_image_sizes.append((val[0], val[1]))# original_image_sizes = [img.shape[-2:] for img in images]images, targets = self.transform(images, targets)  # 对图像进行预处理# print(images.tensors.shape)features = self.backbone(images.tensors)  # 将图像输入backbone得到特征图if isinstance(features, torch.Tensor):  # 若只在一层特征层上预测，将feature放入有序字典中，并编号为‘0’features = OrderedDict([('0', features)])  # 若在多层特征层上预测，传入的就是一个有序字典# 将特征层以及标注target信息传入rpn中# proposals: List[Tensor], Tensor_shape: [num_proposals, 4],# 每个proposals是绝对坐标，且为(x1, y1, x2, y2)格式proposals, proposal_losses = self.rpn(images, features, targets)# 将rpn生成的数据以及标注target信息传入fast rcnn后半部分detections, detector_losses = self.roi_heads(features, proposals, images.image_sizes, targets)# 对网络的预测结果进行后处理（主要将bboxes还原到原图像尺度上）detections = self.transform.postprocess(detections, images.image_sizes, original_image_sizes)losses = {}losses.update(detector_losses)losses.update(proposal_losses)if torch.jit.is_scripting():if not self._has_warned:warnings.warn("RCNN always returns a (Losses, Detections) tuple in scripting")self._has_warned = Truereturn losses, detectionselse:return self.eager_outputs(losses, detections)

首先增加一些容错机制，保住输入数据格式符合模型要求，然后将images和targets输入transforms中进行数据格式的预处理；然后将images输入到backbone中，得到特征图features；将features，images，targets输入rpn网络中，得到proposals和proposals_loss;然后将proposals，images，features等输入到roi_heads得到detections和detector_loss;如果在训练模式下，则返回loss（proposals_loss和detection_loss），在推理模式下，则返回detections。