async def async_test_bboxes(self,
x,
img_metas,
proposals,
rcnn_test_cfg,
rescale=False,
bbox_semaphore=None,
global_lock=None):
"""Asynchronized test for box head without augmentation."""
rois = bbox2roi(proposals)
roi_feats = self.bbox_roi_extractor(
x[:len(self.bbox_roi_extractor.featmap_strides)], rois)
if self.with_shared_head:
roi_feats = self.shared_head(roi_feats)
sleep_interval = rcnn_test_cfg.get('async_sleep_interval', 0.017)
async with completed(__name__,
'bbox_head_forward',
sleep_interval=sleep_interval):
cls_score, bbox_pred = self.bbox_head(roi_feats)
img_shape = img_metas[0]['img_shape']
scale_factor = img_metas[0]['scale_factor']
det_bboxes, det_labels = self.bbox_head.get_bboxes(
rois,
cls_score,
bbox_pred,
img_shape,
scale_factor,
rescale=rescale,
cfg=rcnn_test_cfg)
return det_bboxes, det_labels
def _mask_forward_train(self, x, sampling_results, bbox_feats, gt_masks,
img_metas):
"""Run forward function and calculate loss for mask head in
training."""
if not self.share_roi_extractor:
pos_rois = bbox2roi([res.pos_bboxes for res in sampling_results])
mask_results = self._mask_forward(x, pos_rois)
else:
pos_inds = []
device = bbox_feats.device
for res in sampling_results:
pos_inds.append(
torch.ones(
res.pos_bboxes.shape[0],
device=device,
dtype=torch.uint8))
pos_inds.append(
torch.zeros(
res.neg_bboxes.shape[0],
device=device,
dtype=torch.uint8))
pos_inds = torch.cat(pos_inds)
mask_results = self._mask_forward(
x, pos_inds=pos_inds, bbox_feats=bbox_feats)
mask_targets = self.mask_head.get_targets(sampling_results, gt_masks,
self.train_cfg)
pos_labels = torch.cat([res.pos_gt_labels for res in sampling_results])
loss_mask = self.mask_head.loss(mask_results['mask_pred'],
mask_targets, pos_labels)
mask_results.update(loss_mask=loss_mask, mask_targets=mask_targets)
return mask_results
def aug_test_bboxes(self, feats, img_metas, proposal_list, rcnn_test_cfg):
"""Test det bboxes with test time augmentation."""
aug_bboxes = []
aug_scores = []
for x, img_meta in zip(feats, img_metas):
# only one image in the batch
img_shape = img_meta[0]['img_shape']
scale_factor = img_meta[0]['scale_factor']
flip = img_meta[0]['flip']
flip_direction = img_meta[0]['flip_direction']
# TODO more flexible
proposals = bbox_mapping(proposal_list[0][:, :4], img_shape,
scale_factor, flip, flip_direction)
rois = bbox2roi([proposals])
bbox_results = self._bbox_forward(x, rois)
bboxes, scores = self.bbox_head.get_bboxes(
rois,
bbox_results['cls_score'],
bbox_results['bbox_pred'],
img_shape,
scale_factor,
rescale=False,
cfg=None)
aug_bboxes.append(bboxes)
aug_scores.append(scores)
# after merging, bboxes will be rescaled to the original image size
merged_bboxes, merged_scores = merge_aug_bboxes(
aug_bboxes, aug_scores, img_metas, rcnn_test_cfg)
det_bboxes, det_labels = multiclass_nms(merged_bboxes, merged_scores,
rcnn_test_cfg.score_thr,
rcnn_test_cfg.nms,
rcnn_test_cfg.max_per_img)
return det_bboxes, det_labels
def aug_test_mask(self, feats, img_metas, det_bboxes, det_labels):
"""Test for mask head with test time augmentation."""
if det_bboxes.shape[0] == 0:
segm_result = [[] for _ in range(self.mask_head.num_classes)]
else:
aug_masks = []
for x, img_meta in zip(feats, img_metas):
img_shape = img_meta[0]['img_shape']
scale_factor = img_meta[0]['scale_factor']
flip = img_meta[0]['flip']
flip_direction = img_meta[0]['flip_direction']
_bboxes = bbox_mapping(det_bboxes[:, :4], img_shape,
scale_factor, flip, flip_direction)
mask_rois = bbox2roi([_bboxes])
mask_results = self._mask_forward(x, mask_rois)
# convert to numpy array to save memory
aug_masks.append(
mask_results['mask_pred'].sigmoid().cpu().numpy())
merged_masks = merge_aug_masks(aug_masks, img_metas, self.test_cfg)
ori_shape = img_metas[0][0]['ori_shape']
segm_result = self.mask_head.get_seg_masks(merged_masks,
det_bboxes,
det_labels,
self.test_cfg,
ori_shape,
scale_factor=1.0,
rescale=False)
return segm_result
async def async_test_mask(self,
x,
img_metas,
det_bboxes,
det_labels,
rescale=False,
mask_test_cfg=None):
"""Asynchronized test for mask head without augmentation."""
# image shape of the first image in the batch (only one)
ori_shape = img_metas[0]['ori_shape']
scale_factor = img_metas[0]['scale_factor']
if det_bboxes.shape[0] == 0:
segm_result = [[] for _ in range(self.mask_head.num_classes)]
else:
_bboxes = (det_bboxes[:, :4] *
scale_factor if rescale else det_bboxes)
mask_rois = bbox2roi([_bboxes])
mask_feats = self.mask_roi_extractor(
x[:len(self.mask_roi_extractor.featmap_strides)],
mask_rois)
if self.with_shared_head:
mask_feats = self.shared_head(mask_feats)
if mask_test_cfg and mask_test_cfg.get('async_sleep_interval'):
sleep_interval = mask_test_cfg['async_sleep_interval']
else:
sleep_interval = 0.035
async with completed(__name__,
'mask_head_forward',
sleep_interval=sleep_interval):
mask_pred = self.mask_head(mask_feats)
segm_result = self.mask_head.get_seg_masks(
mask_pred, _bboxes, det_labels, self.test_cfg, ori_shape,
scale_factor, rescale)
return segm_result
def _bbox_forward_train(self, x, sampling_results, gt_bboxes, gt_labels,
img_metas):
"""Run forward function and calculate loss for box head in training."""
rois = bbox2roi([res.bboxes for res in sampling_results])
bbox_results = self._bbox_forward(x, rois) # head网络原始输出结果
# 对应的基于rois的变换targets，用于算loss
bbox_targets = self.bbox_head.get_targets(sampling_results, gt_bboxes,
gt_labels, self.train_cfg)
loss_bbox = self.bbox_head.loss(bbox_results['cls_score'],
bbox_results['bbox_pred'], rois,
*bbox_targets)
bbox_results.update(loss_bbox=loss_bbox)
return bbox_results
def forward_dummy(self, x, proposals):
"""Dummy forward function."""
# bbox head
outs = ()
rois = bbox2roi([proposals])
if self.with_bbox:
bbox_results = self._bbox_forward(x, rois)
outs = outs + (bbox_results['cls_score'],
bbox_results['bbox_pred'])
# mask head
if self.with_mask:
mask_rois = rois[:100]
mask_results = self._mask_forward(x, mask_rois)
outs = outs + (mask_results['mask_pred'], )
return outs
def simple_test_mask(self,
x,
img_metas,
det_bboxes,
det_labels,
rescale=False):
"""Simple test for mask head without augmentation."""
# image shapes of images in the batch
ori_shapes = tuple(meta['ori_shape'] for meta in img_metas)
scale_factors = tuple(meta['scale_factor'] for meta in img_metas)
num_imgs = len(det_bboxes)
if all(det_bbox.shape[0] == 0 for det_bbox in det_bboxes):
segm_results = [[[] for _ in range(self.mask_head.num_classes)]
for _ in range(num_imgs)]
else:
# if det_bboxes is rescaled to the original image size, we need to
# rescale it back to the testing scale to obtain RoIs.
if rescale and not isinstance(scale_factors[0], float):
scale_factors = [
torch.from_numpy(scale_factor).to(det_bboxes[0].device)
for scale_factor in scale_factors
]
_bboxes = [
det_bboxes[i][:, :4] *
scale_factors[i] if rescale else det_bboxes[i][:, :4]
for i in range(len(det_bboxes))
]
mask_rois = bbox2roi(_bboxes)
mask_results = self._mask_forward(x, mask_rois)
mask_pred = mask_results['mask_pred']
# split batch mask prediction back to each image
num_mask_roi_per_img = [len(det_bbox) for det_bbox in det_bboxes]
mask_preds = mask_pred.split(num_mask_roi_per_img, 0)
# apply mask post-processing to each image individually
segm_results = []
for i in range(num_imgs):
if det_bboxes[i].shape[0] == 0:
segm_results.append(
[[] for _ in range(self.mask_head.num_classes)])
else:
segm_result = self.mask_head.get_seg_masks(
mask_preds[i], _bboxes[i], det_labels[i],
self.test_cfg, ori_shapes[i], scale_factors[i],
rescale)
segm_results.append(segm_result)
return segm_results
def simple_test_bboxes(self,
x,
img_metas,
proposals,
rcnn_test_cfg,
rescale=False):
"""Test only det bboxes without augmentation."""
rois = bbox2roi(proposals)
bbox_results = self._bbox_forward(x, rois)
img_shapes = tuple(meta['img_shape'] for meta in img_metas)
scale_factors = tuple(meta['scale_factor'] for meta in img_metas)
# split batch bbox prediction back to each image
cls_score = bbox_results['cls_score']
bbox_pred = bbox_results['bbox_pred']
num_proposals_per_img = tuple(len(p) for p in proposals)
rois = rois.split(num_proposals_per_img, 0)
cls_score = cls_score.split(num_proposals_per_img, 0)
# some detector with_reg is False, bbox_pred will be None
if bbox_pred is not None:
# the bbox prediction of some detectors like SABL is not Tensor
if isinstance(bbox_pred, torch.Tensor):
bbox_pred = bbox_pred.split(num_proposals_per_img, 0)
else:
bbox_pred = self.bbox_head.bbox_pred_split(
bbox_pred, num_proposals_per_img)
else:
bbox_pred = (None, ) * len(proposals)
# apply bbox post-processing to each image individually
det_bboxes = []
det_labels = []
for i in range(len(proposals)):
det_bbox, det_label = self.bbox_head.get_bboxes(rois[i],
cls_score[i],
bbox_pred[i],
img_shapes[i],
scale_factors[i],
rescale=rescale,
cfg=rcnn_test_cfg)
det_bboxes.append(det_bbox)
det_labels.append(det_label)
return det_bboxes, det_labels
def _mask_forward_train(self,
stage,
x,
sampling_results,
gt_masks,
rcnn_train_cfg,
bbox_feats=None):
"""Run forward function and calculate loss for mask head in
training."""
pos_rois = bbox2roi([res.pos_bboxes for res in sampling_results])
mask_results = self._mask_forward(stage, x, pos_rois)
mask_targets = self.mask_head[stage].get_targets(
sampling_results, gt_masks, rcnn_train_cfg)
pos_labels = torch.cat([res.pos_gt_labels for res in sampling_results])
loss_mask = self.mask_head[stage].loss(mask_results['mask_pred'],
mask_targets, pos_labels)
mask_results.update(loss_mask=loss_mask)
return mask_results
def _bbox_forward_train(self, x, sampling_results, gt_bboxes, gt_labels,
img_metas):
num_imgs = len(img_metas)
rois = bbox2roi([res.bboxes for res in sampling_results])
bbox_results = self._bbox_forward(x, rois)
bbox_targets = self.bbox_head.get_targets(sampling_results, gt_bboxes,
gt_labels, self.train_cfg)
# record the `beta_topk`-th smallest target
# `bbox_targets[2]` and `bbox_targets[3]` stand for bbox_targets
# and bbox_weights, respectively
pos_inds = bbox_targets[3][:, 0].nonzero().squeeze(1)
num_pos = len(pos_inds)
cur_target = bbox_targets[2][pos_inds, :2].abs().mean(dim=1)
beta_topk = min(self.train_cfg.dynamic_rcnn.beta_topk * num_imgs,
num_pos)
cur_target = torch.kthvalue(cur_target, beta_topk)[0].item()
self.beta_history.append(cur_target)
loss_bbox = self.bbox_head.loss(bbox_results['cls_score'],
bbox_results['bbox_pred'], rois,
*bbox_targets)
bbox_results.update(loss_bbox=loss_bbox)
return bbox_results
def aug_test(self, features, proposal_list, img_metas, rescale=False):
"""Test with augmentations.
If rescale is False, then returned bboxes and masks will fit the scale
of imgs[0].
"""
rcnn_test_cfg = self.test_cfg
aug_bboxes = []
aug_scores = []
for x, img_meta in zip(features, img_metas):
# only one image in the batch
img_shape = img_meta[0]['img_shape']
scale_factor = img_meta[0]['scale_factor']
flip = img_meta[0]['flip']
flip_direction = img_meta[0]['flip_direction']
proposals = bbox_mapping(proposal_list[0][:, :4], img_shape,
scale_factor, flip, flip_direction)
# "ms" in variable names means multi-stage
ms_scores = []
rois = bbox2roi([proposals])
for i in range(self.num_stages):
bbox_results = self._bbox_forward(i, x, rois)
ms_scores.append(bbox_results['cls_score'])
if i < self.num_stages - 1:
bbox_label = bbox_results['cls_score'][:, :-1].argmax(
dim=1)
rois = self.bbox_head[i].regress_by_class(
rois, bbox_label, bbox_results['bbox_pred'],
img_meta[0])
cls_score = sum(ms_scores) / float(len(ms_scores))
bboxes, scores = self.bbox_head[-1].get_bboxes(
rois,
cls_score,
bbox_results['bbox_pred'],
img_shape,
scale_factor,
rescale=False,
cfg=None)
aug_bboxes.append(bboxes)
aug_scores.append(scores)
# after merging, bboxes will be rescaled to the original image size
merged_bboxes, merged_scores = merge_aug_bboxes(
aug_bboxes, aug_scores, img_metas, rcnn_test_cfg)
det_bboxes, det_labels = multiclass_nms(merged_bboxes, merged_scores,
rcnn_test_cfg.score_thr,
rcnn_test_cfg.nms,
rcnn_test_cfg.max_per_img)
bbox_result = bbox2result(det_bboxes, det_labels,
self.bbox_head[-1].num_classes)
if self.with_mask:
if det_bboxes.shape[0] == 0:
segm_result = [[[]
for _ in range(self.mask_head[-1].num_classes)]
]
else:
aug_masks = []
aug_img_metas = []
for x, img_meta in zip(features, img_metas):
img_shape = img_meta[0]['img_shape']
scale_factor = img_meta[0]['scale_factor']
flip = img_meta[0]['flip']
flip_direction = img_meta[0]['flip_direction']
_bboxes = bbox_mapping(det_bboxes[:, :4], img_shape,
scale_factor, flip, flip_direction)
mask_rois = bbox2roi([_bboxes])
for i in range(self.num_stages):
mask_results = self._mask_forward(i, x, mask_rois)
aug_masks.append(
mask_results['mask_pred'].sigmoid().cpu().numpy())
aug_img_metas.append(img_meta)
merged_masks = merge_aug_masks(aug_masks, aug_img_metas,
self.test_cfg)
ori_shape = img_metas[0][0]['ori_shape']
segm_result = self.mask_head[-1].get_seg_masks(
merged_masks,
det_bboxes,
det_labels,
rcnn_test_cfg,
ori_shape,
scale_factor=1.0,
rescale=False)
return [(bbox_result, segm_result)]
else:
return [bbox_result]
def simple_test(self, x, proposal_list, img_metas, rescale=False):
"""Test without augmentation."""
assert self.with_bbox, 'Bbox head must be implemented.'
num_imgs = len(proposal_list)
img_shapes = tuple(meta['img_shape'] for meta in img_metas)
ori_shapes = tuple(meta['ori_shape'] for meta in img_metas)
scale_factors = tuple(meta['scale_factor'] for meta in img_metas)
# "ms" in variable names means multi-stage
ms_bbox_result = {}
ms_segm_result = {}
ms_scores = []
rcnn_test_cfg = self.test_cfg
rois = bbox2roi(proposal_list)
for i in range(self.num_stages):
bbox_results = self._bbox_forward(i, x, rois)
# split batch bbox prediction back to each image
cls_score = bbox_results['cls_score']
bbox_pred = bbox_results['bbox_pred']
num_proposals_per_img = tuple(
len(proposals) for proposals in proposal_list)
rois = rois.split(num_proposals_per_img, 0)
cls_score = cls_score.split(num_proposals_per_img, 0)
if isinstance(bbox_pred, torch.Tensor):
bbox_pred = bbox_pred.split(num_proposals_per_img, 0)
else:
bbox_pred = self.bbox_head[i].bbox_pred_split(
bbox_pred, num_proposals_per_img)
ms_scores.append(cls_score)
if i < self.num_stages - 1:
bbox_label = [s[:, :-1].argmax(dim=1) for s in cls_score]
rois = torch.cat([
self.bbox_head[i].regress_by_class(rois[j], bbox_label[j],
bbox_pred[j],
img_metas[j])
for j in range(num_imgs)
])
# average scores of each image by stages
cls_score = [
sum([score[i] for score in ms_scores]) / float(len(ms_scores))
for i in range(num_imgs)
]
# apply bbox post-processing to each image individually
det_bboxes = []
det_labels = []
for i in range(num_imgs):
det_bbox, det_label = self.bbox_head[-1].get_bboxes(
rois[i],
cls_score[i],
bbox_pred[i],
img_shapes[i],
scale_factors[i],
rescale=rescale,
cfg=rcnn_test_cfg)
det_bboxes.append(det_bbox)
det_labels.append(det_label)
bbox_results = [
bbox2result(det_bboxes[i], det_labels[i],
self.bbox_head[-1].num_classes)
for i in range(num_imgs)
]
ms_bbox_result['ensemble'] = bbox_results
if self.with_mask:
if all(det_bbox.shape[0] == 0 for det_bbox in det_bboxes):
mask_classes = self.mask_head[-1].num_classes
segm_results = [[[] for _ in range(mask_classes)]
for _ in range(num_imgs)]
else:
if rescale and not isinstance(scale_factors[0], float):
scale_factors = [
torch.from_numpy(scale_factor).to(det_bboxes[0].device)
for scale_factor in scale_factors
]
_bboxes = [
det_bboxes[i][:, :4] *
scale_factors[i] if rescale else det_bboxes[i][:, :4]
for i in range(len(det_bboxes))
]
mask_rois = bbox2roi(_bboxes)
num_mask_rois_per_img = tuple(
_bbox.size(0) for _bbox in _bboxes)
aug_masks = []
for i in range(self.num_stages):
mask_results = self._mask_forward(i, x, mask_rois)
mask_pred = mask_results['mask_pred']
# split batch mask prediction back to each image
mask_pred = mask_pred.split(num_mask_rois_per_img, 0)
aug_masks.append(
[m.sigmoid().cpu().numpy() for m in mask_pred])
# apply mask post-processing to each image individually
segm_results = []
for i in range(num_imgs):
if det_bboxes[i].shape[0] == 0:
segm_results.append(
[[]
for _ in range(self.mask_head[-1].num_classes)])
else:
aug_mask = [mask[i] for mask in aug_masks]
merged_masks = merge_aug_masks(
aug_mask, [[img_metas[i]]] * self.num_stages,
rcnn_test_cfg)
segm_result = self.mask_head[-1].get_seg_masks(
merged_masks, _bboxes[i], det_labels[i],
rcnn_test_cfg, ori_shapes[i], scale_factors[i],
rescale)
segm_results.append(segm_result)
ms_segm_result['ensemble'] = segm_results
if self.with_mask:
results = list(
zip(ms_bbox_result['ensemble'], ms_segm_result['ensemble']))
else:
results = ms_bbox_result['ensemble']
return results