EPNet代码（2）

class RPN(nn.Module):
    def __init__(self, use_xyz = True, mode = 'TRAIN'):
        super().__init__()
        self.training_mode = (mode == 'TRAIN')

        # MODEL = importlib.import_module(cfg.RPN.BACKBONE)
        # self.backbone_net = MODEL.get_model(input_channels=int(cfg.RPN.USE_INTENSITY), use_xyz=use_xyz)
        input_channels = int(cfg.RPN.USE_INTENSITY) + 3 * int(cfg.RPN.USE_RGB)
        if cfg.RPN.BACKBONE == 'pointnet2_msg':
            self.backbone_net = Pointnet2MSG(input_channels =input_channels, use_xyz = use_xyz)

        # classification branch
        cls_layers = []
        pre_channel = cfg.RPN.FP_MLPS[0][-1]
        for k in range(0, cfg.RPN.CLS_FC.__len__()):
            cls_layers.append(pt_utils.Conv1d(pre_channel, cfg.RPN.CLS_FC[k], bn = cfg.RPN.USE_BN))
            pre_channel = cfg.RPN.CLS_FC[k]
        cls_layers.append(pt_utils.Conv1d(pre_channel, 1, activation = None))
        if cfg.RPN.DP_RATIO >= 0:
            cls_layers.insert(1, nn.Dropout(cfg.RPN.DP_RATIO))
        self.rpn_cls_layer = nn.Sequential(*cls_layers)

        # regression branch回归分支
        per_loc_bin_num = int(cfg.RPN.LOC_SCOPE / cfg.RPN.LOC_BIN_SIZE) * 2
        if cfg.RPN.LOC_XZ_FINE:
            reg_channel = per_loc_bin_num * 4 + cfg.RPN.NUM_HEAD_BIN * 2 + 3
        else:
            reg_channel = per_loc_bin_num * 2 + cfg.RPN.NUM_HEAD_BIN * 2 + 3
        reg_channel += 1  # reg y

        reg_layers = []
        pre_channel = cfg.RPN.FP_MLPS[0][-1]
        for k in range(0, cfg.RPN.REG_FC.__len__()):
            reg_layers.append(pt_utils.Conv1d(pre_channel, cfg.RPN.REG_FC[k], bn = cfg.RPN.USE_BN))##pre_channel=76
            pre_channel = cfg.RPN.REG_FC[k]
        reg_layers.append(pt_utils.Conv1d(pre_channel, reg_channel, activation = None))
        if cfg.RPN.DP_RATIO >= 0:
            reg_layers.insert(1, nn.Dropout(cfg.RPN.DP_RATIO))
        self.rpn_reg_layer = nn.Sequential(*reg_layers)

        if cfg.RPN.LOSS_CLS == 'DiceLoss':
            self.rpn_cls_loss_func = loss_utils.DiceLoss(ignore_target = -1)
        elif cfg.RPN.LOSS_CLS == 'SigmoidFocalLoss':
            self.rpn_cls_loss_func = loss_utils.SigmoidFocalClassificationLoss(alpha = cfg.RPN.FOCAL_ALPHA[0],
                                                                               gamma = cfg.RPN.FOCAL_GAMMA)
        elif cfg.RPN.LOSS_CLS == 'BinaryCrossEntropy':
            self.rpn_cls_loss_func = F.binary_cross_entropy
        else:
            raise NotImplementedError

        self.proposal_layer = ProposalLayer(mode = mode)
        self.init_weights()

    def init_weights(self):
        if cfg.RPN.LOSS_CLS in ['SigmoidFocalLoss']:
            pi = 0.01
            nn.init.constant_(self.rpn_cls_layer[2].conv.bias, -np.log((1 - pi) / pi))

        nn.init.normal_(self.rpn_reg_layer[-1].conv.weight, mean = 0, std = 0.001)###初始化网络权重

    def forward(self, input_data):
        """
        :param input_data: dict (point_cloud)
        :return:
        """
        pts_input = input_data['pts_input']
        if cfg.LI_FUSION.ENABLED:
            img_input = input_data['img']
            xy_input = input_data['pts_origin_xy']
            backbone_xyz, backbone_features = self.backbone_net(pts_input, img_input, xy_input)  # (B, N, 3), (B, C, N)
        else:
            backbone_xyz, backbone_features = self.backbone_net(pts_input)  # (B, N, 3), (B, C, N)


        rpn_cls = self.rpn_cls_layer(backbone_features).transpose(1, 2).contiguous()  # (B, N, 1)
        rpn_reg = self.rpn_reg_layer(backbone_features).transpose(1, 2).contiguous()  # (B, N, 76)

        ret_dict = { 'rpn_cls': rpn_cls, 'rpn_reg': rpn_reg,
                     'backbone_xyz': backbone_xyz, 'backbone_features': backbone_features }

        return ret_dict

作者是用PointRCNN代码修改的   也是用76维去回归7维的候选框这部分跟PointRCNN代码一样。

：初始化网络参数

    def forward(self, input_data):
        """
        :param input_data: dict (point_cloud)
        :return:
        """
        pts_input = input_data['pts_input']
        if cfg.LI_FUSION.ENABLED:
            img_input = input_data['img']
            xy_input = input_data['pts_origin_xy']
            backbone_xyz, backbone_features = self.backbone_net(pts_input, img_input, xy_input)  # (B, N, 3), (B, C, N)
        else:
            backbone_xyz, backbone_features = self.backbone_net(pts_input)  # (B, N, 3), (B, C, N)


        rpn_cls = self.rpn_cls_layer(backbone_features).transpose(1, 2).contiguous()  # (B, N, 1)
        rpn_reg = self.rpn_reg_layer(backbone_features).transpose(1, 2).contiguous()  # (B, N, 76)

        ret_dict = { 'rpn_cls': rpn_cls, 'rpn_reg': rpn_reg,
                     'backbone_xyz': backbone_xyz, 'backbone_features': backbone_features }

        return ret_dict

 然后使用backbone_net去提取特征，在得到分类和回归的结果。其中经过分类得到的分数和回归会被送入候选框生成里面预测候选框。