Source code for mmrotate.models.dense_heads.odm_refine_head

# Copyright (c) OpenMMLab. All rights reserved.
import torch.nn as nn
from mmcv.cnn import ConvModule
from mmcv.runner import force_fp32

from ..builder import ROTATED_HEADS
from ..utils import ORConv2d, RotationInvariantPooling
from .rotated_retina_head import RotatedRetinaHead


@ROTATED_HEADS.register_module()
class ODMRefineHead(RotatedRetinaHead):
    """Rotated Anchor-based refine head. It's a part of the Oriented Detection
    Module (ODM), which produces orientation-sensitive features for
    classification and orientation-invariant features for localization.

    Args:
        num_classes (int): Number of categories excluding the background
            category.
        in_channels (int): Number of channels in the input feature map.
        stacked_convs (int, optional): Number of stacked convolutions.
        conv_cfg (dict, optional): Config dict for convolution layer.
            Default: None.
        norm_cfg (dict, optional): Config dict for normalization layer.
            Default: None.
        anchor_generator (dict): Config dict for anchor generator
        init_cfg (dict or list[dict], optional): Initialization config dict.
    """  # noqa: W605

    def __init__(self,
                 num_classes,
                 in_channels,
                 stacked_convs=2,
                 conv_cfg=None,
                 norm_cfg=None,
                 anchor_generator=dict(
                     type='PseudoAnchorGenerator',
                     strides=[8, 16, 32, 64, 128]),
                 init_cfg=dict(
                     type='Normal',
                     layer='Conv2d',
                     std=0.01,
                     override=dict(
                         type='Normal',
                         name='odm_cls',
                         std=0.01,
                         bias_prob=0.01)),
                 **kwargs):
        self.bboxes_as_anchors = None
        self.stacked_convs = stacked_convs
        self.conv_cfg = conv_cfg
        self.norm_cfg = norm_cfg
        super(ODMRefineHead, self).__init__(
            num_classes,
            in_channels,
            stacked_convs=2,
            anchor_generator=anchor_generator,
            init_cfg=init_cfg,
            **kwargs)

    def _init_layers(self):
        """Initialize layers of the head."""

        self.or_conv = ORConv2d(
            self.feat_channels,
            int(self.feat_channels / 8),
            kernel_size=3,
            padding=1,
            arf_config=(1, 8))
        self.or_pool = RotationInvariantPooling(256, 8)
        self.cls_convs = nn.ModuleList()
        self.reg_convs = nn.ModuleList()
        for i in range(self.stacked_convs):
            chn = int(self.feat_channels / 8) if i == 0 else self.feat_channels
            self.reg_convs.append(
                ConvModule(
                    self.feat_channels,
                    self.feat_channels,
                    3,
                    stride=1,
                    padding=1,
                    conv_cfg=self.conv_cfg,
                    norm_cfg=self.norm_cfg))
            self.cls_convs.append(
                ConvModule(
                    chn,
                    self.feat_channels,
                    3,
                    stride=1,
                    padding=1,
                    conv_cfg=self.conv_cfg,
                    norm_cfg=self.norm_cfg))
        self.odm_cls = nn.Conv2d(
            self.feat_channels,
            self.num_anchors * self.cls_out_channels,
            3,
            padding=1)
        self.odm_reg = nn.Conv2d(
            self.feat_channels, self.num_anchors * 5, 3, padding=1)

    def forward_single(self, x):
        """Forward feature of a single scale level.

        Args:
            x (torch.Tensor): Features of a single scale level.

        Returns:
            tuple (torch.Tensor):

                - cls_score (torch.Tensor): Cls scores for a single scale \
                    level the channels number is num_anchors * num_classes.
                - bbox_pred (torch.Tensor): Box energies / deltas for a \
                    single scale level, the channels number is num_anchors * 4.
        """
        or_feat = self.or_conv(x)
        reg_feat = or_feat
        cls_feat = self.or_pool(or_feat)
        for cls_conv in self.cls_convs:
            cls_feat = cls_conv(cls_feat)
        for reg_conv in self.reg_convs:
            reg_feat = reg_conv(reg_feat)
        cls_score = self.odm_cls(cls_feat)
        bbox_pred = self.odm_reg(reg_feat)
        return cls_score, bbox_pred

    def get_anchors(self, featmap_sizes, img_metas, device='cuda'):
        """Get anchors according to feature map sizes.

        Args:
            featmap_sizes (list[tuple]): Multi-level feature map sizes.
            img_metas (list[dict]): Image meta info.
            bboxes_as_anchors (list[list[Tensor]]) bboxes of levels of images.
                before further regression just like anchors.
            device (torch.device | str): Device for returned tensors

        Returns:
            tuple (list[Tensor]):

                - anchor_list (list[Tensor]): Anchors of each image
                - valid_flag_list (list[Tensor]): Valid flags of each image
        """
        anchor_list = [[
            bboxes_img_lvl.clone().detach() for bboxes_img_lvl in bboxes_img
        ] for bboxes_img in self.bboxes_as_anchors]

        # for each image, we compute valid flags of multi level anchors
        valid_flag_list = []
        for img_id, img_meta in enumerate(img_metas):
            multi_level_flags = self.anchor_generator.valid_flags(
                featmap_sizes, img_meta['pad_shape'], device)
            valid_flag_list.append(multi_level_flags)

        return anchor_list, valid_flag_list

    @force_fp32(apply_to=('cls_scores', 'bbox_preds'))
    def loss(self,
             cls_scores,
             bbox_preds,
             gt_bboxes,
             gt_labels,
             img_metas,
             rois=None,
             gt_bboxes_ignore=None):
        """Loss function of ODMRefineHead."""
        assert rois is not None
        self.bboxes_as_anchors = rois
        return super(ODMRefineHead, self).loss(
            cls_scores=cls_scores,
            bbox_preds=bbox_preds,
            gt_bboxes=gt_bboxes,
            gt_labels=gt_labels,
            img_metas=img_metas,
            gt_bboxes_ignore=gt_bboxes_ignore)

    @force_fp32(apply_to=('cls_scores', 'bbox_preds'))
    def get_bboxes(self,
                   cls_scores,
                   bbox_preds,
                   img_metas,
                   cfg=None,
                   rescale=False,
                   rois=None):
        """Transform network output for a batch into labeled boxes.

        Args:
            cls_scores (list[Tensor]): Box scores for each scale level
                Has shape (N, num_anchors * num_classes, H, W)
            bbox_preds (list[Tensor]): Box energies / deltas for each scale
                level with shape (N, num_anchors * 5, H, W)
            img_metas (list[dict]): size / scale info for each image
            cfg (mmcv.Config): test / postprocessing configuration
            rois (list[list[Tensor]]): input rbboxes of each level of
            each image. rois output by former stages and are to be refined
            rescale (bool): if True, return boxes in original image space

        Returns:
            list[tuple[Tensor, Tensor]]: each item in result_list is 2-tuple.
                The first item is an (n, 6) tensor, where the first 5 columns
                are bounding box positions (xc, yc, w, h, a) and the
                6-th column is a score between 0 and 1. The second item is a
                (n,) tensor where each item is the class index of the
                corresponding box.
        """
        num_levels = len(cls_scores)
        assert len(cls_scores) == len(bbox_preds)
        assert rois is not None

        result_list = []

        for img_id, _ in enumerate(img_metas):
            cls_score_list = [
                cls_scores[i][img_id].detach() for i in range(num_levels)
            ]
            bbox_pred_list = [
                bbox_preds[i][img_id].detach() for i in range(num_levels)
            ]
            img_shape = img_metas[img_id]['img_shape']
            scale_factor = img_metas[img_id]['scale_factor']
            proposals = self._get_bboxes_single(cls_score_list, bbox_pred_list,
                                                rois[img_id], img_shape,
                                                scale_factor, cfg, rescale)
            result_list.append(proposals)
        return result_list