Mask Scoring RCNN 解读

Introduction

相比于Mask RCNN 多了一条head计算MaskIOU

 

mask_head/loss.py

rle_for_fullarea 计算的是ground truth mask的面积

rle_for_box_area 计算的是proposal cropped mask的面积 

mask_ratio 算的是我predict的proposal 里面的mask相比与ground truth的比例

mask_targets是我prediction proposal 里面的mask 不是ground truth mask 它会除以mask ratio计算得到ground truth的mask area

 

# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
import torch
from torch.nn import functional as F
import pycocotools.mask as mask_util

from maskrcnn_benchmark.layers import smooth_l1_loss
from maskrcnn_benchmark.modeling.matcher import Matcher
from maskrcnn_benchmark.structures.boxlist_ops import boxlist_iou
from maskrcnn_benchmark.modeling.utils import cat
import numpy as np
import cv2

def project_masks_on_boxes(segmentation_masks, proposals, discretization_size, maskiou_on):
    """
    Given segmentation masks and the bounding boxes corresponding
    to the location of the masks in the image, this function
    crops and resizes the masks in the position defined by the
    boxes. This prepares the masks for them to be fed to the
    loss computation as the targets. If use maskiou head, we will compute the maskiou target here.

    Arguments:
        segmentation_masks: an instance of SegmentationMask
        proposals: an instance of BoxList
    """
    masks = []
    mask_ratios = []
    M = discretization_size
    device = proposals.bbox.device
    proposals = proposals.convert("xyxy")
    assert segmentation_masks.size == proposals.size, "{}, {}".format(
        segmentation_masks, proposals
    )
    # TODO put the proposals on the CPU, as the representation for the
    # masks is not efficient GPU-wise (possibly several small tensors for
    # representing a single instance mask)
    proposals = proposals.bbox.to(torch.device("cpu"))
    for segmentation_mask, proposal in zip(segmentation_masks, proposals):
        # crop the masks, resize them to the desired resolution and
        # then convert them to the tensor representation,
        # instead of the list representation that was used
        cropped_mask = segmentation_mask.crop(proposal)
        scaled_mask = cropped_mask.resize((M, M))
        mask = scaled_mask.convert(mode="mask")
        masks.append(mask)
        if maskiou_on:
            
            x1 = int(proposal[0])
            y1 = int(proposal[1])
            x2 = int(proposal[2]) + 1
            y2 = int(proposal[3]) + 1
            for poly_ in segmentation_mask.polygons:
                poly = np.array(poly_, dtype=np.float32)
                x1 = np.minimum(x1, poly[0::2].min())
                x2 = np.maximum(x2, poly[0::2].max())
                y1 = np.minimum(y1, poly[1::2].min())
                y2 = np.maximum(y2, poly[1::2].max())
            img_h = segmentation_mask.size[1]
            img_w = segmentation_mask.size[0]
            x1 = np.maximum(x1, 0)
            x2 = np.minimum(x2, img_w-1)
            y1 = np.maximum(y1, 0)
            y2 = np.minimum(y2, img_h-1)
            segmentation_mask_for_maskratio =  segmentation_mask.crop([x1, y1, x2, y2])
            ''' 
            #type 1
            gt_img_mask = segmentation_mask_for_maskratio.convert(mode='mask')    
            gt_img_mask_area = gt_img_mask.sum().float()
            gt_box_mask = gt_img_mask[int(proposal[1]-y1):int(proposal[3]-y1)+1, int(proposal[0]-x1):int(proposal[2]-x1)+1]
            gt_box_mask_area = gt_box_mask.sum().float()
            mask_ratio = gt_box_mask_area / gt_img_mask_area
            '''
            #type 2
            rle_for_fullarea = mask_util.frPyObjects([p.numpy() for p in segmentation_mask_for_maskratio.polygons], y2-y1, x2-x1)
            full_area = torch.tensor(mask_util.area(rle_for_fullarea).sum().astype(float))
            rle_for_box_area = mask_util.frPyObjects([p.numpy() for p in cropped_mask.polygons], proposal[3]-proposal[1], proposal[2]-proposal[0])
            box_area = torch.tensor(mask_util.area(rle_for_box_area).sum().astype(float))
            mask_ratio = box_area / full_area

            mask_ratios.append(mask_ratio)
    if maskiou_on:
        mask_ratios = torch.stack(mask_ratios, dim=0).to(device, dtype=torch.float32)
    else:
        mask_ratios = None
    if len(masks) == 0:
        return torch.empty(0, dtype=torch.float32, device=device), torch.empty(0, dtype=torch.float32, device=device)
    return torch.stack(masks, dim=0).to(device, dtype=torch.float32), mask_ratios

class MaskRCNNLossComputation(object):
    def __init__(self, proposal_matcher, discretization_size, maskiou_on):
        """
        Arguments:
            proposal_matcher (Matcher)
            discretization_size (int)
        """
        self.proposal_matcher = proposal_matcher
        self.discretization_size = discretization_size
        self.maskiou_on = maskiou_on

    def match_targets_to_proposals(self, proposal, target):
        match_quality_matrix = boxlist_iou(target, proposal)
        matched_idxs = self.proposal_matcher(match_quality_matrix)
        # Mask RCNN needs "labels" and "masks "fields for creating the targets
        target = target.copy_with_fields(["labels", "masks"])
        # get the targets corresponding GT for each proposal
        # NB: need to clamp the indices because we can have a single
        # GT in the image, and matched_idxs can be -2, which goes
        # out of bounds
        matched_targets = target[matched_idxs.clamp(min=0)]
        matched_targets.add_field("matched_idxs", matched_idxs)
        return matched_targets

    def prepare_targets(self, proposals, targets):
        labels = []
        masks = []
        mask_ratios = []
        for proposals_per_image, targets_per_image in zip(proposals, targets):
            matched_targets = self.match_targets_to_proposals(
                proposals_per_image, targets_per_image
            )
            matched_idxs = matched_targets.get_field("matched_idxs")

            labels_per_image = matched_targets.get_field("labels")
            labels_per_image = labels_per_image.to(dtype=torch.int64)

            # this can probably be removed, but is left here for clarity
            # and completeness
            neg_inds = matched_idxs == Matcher.BELOW_LOW_THRESHOLD
            labels_per_image[neg_inds] = 0

            # mask scores are only computed on positive samples
            positive_inds = torch.nonzero(labels_per_image > 0).squeeze(1)
            
            segmentation_masks = matched_targets.get_field("masks")
            segmentation_masks = segmentation_masks[positive_inds]
            
            positive_proposals = proposals_per_image[positive_inds]

            masks_per_image, mask_ratios_per_image = project_masks_on_boxes(
                segmentation_masks, positive_proposals, self.discretization_size, self.maskiou_on
            )
            
            labels.append(labels_per_image)
            masks.append(masks_per_image)
            mask_ratios.append(mask_ratios_per_image)
        return labels, masks, mask_ratios

    def __call__(self, proposals, mask_logits, targets):
        """
        Arguments:
            proposals (list[BoxList])
            mask_logits (Tensor)
            targets (list[BoxList])

        Return:
            mask_loss (Tensor): scalar tensor containing the loss
            If we use maskiou head, we will return extra feature for maskiou head.
        """
        labels, mask_targets, mask_ratios = self.prepare_targets(proposals, targets)

        labels = cat(labels, dim=0)
        mask_targets = cat(mask_targets, dim=0)

        positive_inds = torch.nonzero(labels > 0).squeeze(1)
        labels_pos = labels[positive_inds]

        # torch.mean (in binary_cross_entropy_with_logits) doesn't
        # accept empty tensors, so handle it separately
        if mask_targets.numel() == 0:
            if not self.maskiou_on:
                return mask_logits.sum() * 0
            else:
                selected_index = torch.arange(mask_logits.shape[0], device=labels.device)
                selected_mask = mask_logits[selected_index, labels]
                mask_num, mask_h, mask_w = selected_mask.shape
                selected_mask = selected_mask.reshape(mask_num, 1, mask_h, mask_w)
                return mask_logits.sum() * 0, selected_mask, labels, None
        
        if self.maskiou_on:
            mask_ratios = cat(mask_ratios, dim=0)
            value_eps = 1e-10 * torch.ones(mask_targets.shape[0], device=labels.device)
            mask_ratios = torch.max(mask_ratios, value_eps)
            pred_masks = mask_logits[positive_inds, labels_pos]
            pred_masks[:] = pred_masks > 0
            mask_targets_full_area = mask_targets.sum(dim=[1,2]) / mask_ratios
            mask_ovr = pred_masks * mask_targets
            mask_ovr_area = mask_ovr.sum(dim=[1,2])
            mask_union_area = pred_masks.sum(dim=[1,2]) + mask_targets_full_area - mask_ovr_area
            value_1 = torch.ones(pred_masks.shape[0], device=labels.device)
            value_0 = torch.zeros(pred_masks.shape[0], device=labels.device)
            mask_union_area = torch.max(mask_union_area, value_1)
            mask_ovr_area = torch.max(mask_ovr_area, value_0)
            maskiou_targets = mask_ovr_area / mask_union_area
            
        mask_loss = F.binary_cross_entropy_with_logits(
            mask_logits[positive_inds, labels_pos], mask_targets
        )
        if not self.maskiou_on:
            return mask_loss
        else:
            selected_index = torch.arange(mask_logits.shape[0], device=labels.device)
            selected_mask = mask_logits[selected_index, labels]
            mask_num, mask_h, mask_w = selected_mask.shape
            selected_mask = selected_mask.reshape(mask_num, 1, mask_h, mask_w)
            selected_mask = selected_mask.sigmoid()
            return mask_loss, selected_mask, labels, maskiou_targets


def make_roi_mask_loss_evaluator(cfg):
    matcher = Matcher(
        cfg.MODEL.ROI_HEADS.FG_IOU_THRESHOLD,
        cfg.MODEL.ROI_HEADS.BG_IOU_THRESHOLD,
        allow_low_quality_matches=False,
    )

    loss_evaluator = MaskRCNNLossComputation(
        matcher, cfg.MODEL.ROI_MASK_HEAD.RESOLUTION, cfg.MODEL.MASKIOU_ON
    )

    return loss_evaluator

 

 

Reference

http://openaccess.thecvf.com/content_CVPR_2019/papers/Huang_Mask_Scoring_R-CNN_CVPR_2019_paper.pdf

https://github.com/zjhuang22/maskscoring_rcnn