def boxlist_box_voting(top_boxlist,
all_boxlist,
thresh,
scoring_method='ID',
beta=1.0,
score_field="scores"):
if thresh <= 0:
return top_boxlist
mode = top_boxlist.mode
top_boxes = top_boxlist.convert("xyxy").bbox.cpu()
all_boxes = all_boxlist.convert("xyxy").bbox.cpu()
top_score = top_boxlist.get_field(score_field).cpu()
all_score = all_boxlist.get_field(score_field).cpu()
top_dets = np.hstack((top_boxes, top_score[:,
np.newaxis])).astype(np.float32,
copy=False)
all_dets = np.hstack((all_boxes, all_score[:,
np.newaxis])).astype(np.float32,
copy=False)
dets = box_utils.box_voting(top_dets, all_dets, thresh, scoring_method,
beta)
boxlist = BoxList(torch.from_numpy(dets[:, :4]).cuda(),
all_boxlist.size,
mode="xyxy")
boxlist.add_field("scores", torch.from_numpy(dets[:, -1]).cuda())
return boxlist.convert(mode)
def cat_boxlist(bboxes):
"""
Concatenates a list of BoxList (having the same image size) into a
single BoxList
Arguments:
bboxes (list[BoxList])
"""
assert isinstance(bboxes, (list, tuple))
assert all(isinstance(bbox, BoxList) for bbox in bboxes)
size = bboxes[0].size
assert all(bbox.size == size for bbox in bboxes)
mode = bboxes[0].mode
assert all(bbox.mode == mode for bbox in bboxes)
fields = set(bboxes[0].fields())
assert all(set(bbox.fields()) == fields for bbox in bboxes)
cat_boxes = BoxList(_cat([bbox.bbox for bbox in bboxes], dim=0), size,
mode)
for field in fields:
data = _cat([bbox.get_field(field) for bbox in bboxes], dim=0)
cat_boxes.add_field(field, data)
return cat_boxes
def boxlist_soft_nms(boxlist,
sigma=0.5,
overlap_thresh=0.3,
score_thresh=0.001,
method='linear',
score_field="scores"):
"""
Performs non-maximum suppression on a boxlist, with scores specified
in a boxlist field via score_field.
Arguments:
boxlist(BoxList)
nms_thresh (float)
max_proposals (int): if > 0, then only the top max_proposals are kept
after non-maximum suppression
score_field (str)
"""
if overlap_thresh <= 0:
return boxlist
mode = boxlist.mode
boxlist = boxlist.convert("xyxy")
boxes = boxlist.bbox.cpu()
score = boxlist.get_field(score_field).cpu()
dets = np.hstack((boxes, score[:, np.newaxis])).astype(np.float32,
copy=False)
dets, _ = box_utils.soft_nms(dets, sigma, overlap_thresh, score_thresh,
method)
boxlist = BoxList(torch.from_numpy(dets[:, :4]).cuda(),
boxlist.size,
mode="xyxy")
boxlist.add_field("scores", torch.from_numpy(dets[:, -1]).cuda())
return boxlist.convert(mode)
def boxlist_soft_nms(boxlist,
sigma=0.5,
overlap_thresh=0.3,
score_thresh=0.001,
method='linear',
score_field="scores"):
"""
Performs non-maximum suppression on a boxlist, with scores specified
in a boxlist field via score_field.
Arguments:
boxlist(BoxList)
nms_thresh (float)
max_proposals (int): if > 0, then only the top max_proposals are kept
after non-maximum suppression
score_field (str)
"""
if overlap_thresh <= 0:
return boxlist
mode = boxlist.mode
boxlist = boxlist.convert("xyxy")
boxes = boxlist.bbox.cpu()
score = boxlist.get_field(score_field).cpu()
dets, scores, _ = _box_soft_nms(boxes, score, sigma, overlap_thresh,
score_thresh, method)
boxlist = BoxList(dets.cuda(), boxlist.size, mode="xyxy")
boxlist.add_field("scores", scores.cuda())
return boxlist.convert(mode)
def forward_for_single_feature_map(self, anchors, objectness,
box_regression):
"""
Arguments:
anchors: list[BoxList]
objectness: tensor of size N, A, H, W
box_regression: tensor of size N, A * 4, H, W
"""
device = objectness.device
N, A, H, W = objectness.shape
# put in the same format as anchors
objectness = permute_and_flatten(objectness, N, A, 1, H, W).view(N, -1)
objectness = objectness.sigmoid()
box_regression = permute_and_flatten(box_regression, N, A, 4, H, W)
num_anchors = A * H * W
pre_nms_top_n = min(self.pre_nms_top_n, num_anchors)
objectness, topk_idx = objectness.topk(pre_nms_top_n,
dim=1,
sorted=True)
batch_idx = torch.arange(N, device=device)[:, None]
box_regression = box_regression[batch_idx, topk_idx]
image_shapes = [box.size for box in anchors]
concat_anchors = torch.cat([a.bbox for a in anchors], dim=0)
concat_anchors = concat_anchors.reshape(N, -1, 4)[batch_idx, topk_idx]
proposals = self.box_coder.decode(box_regression.view(-1, 4),
concat_anchors.view(-1, 4))
proposals = proposals.view(N, -1, 4)
result = []
for proposal, score, im_shape in zip(proposals, objectness,
image_shapes):
boxlist = BoxList(proposal, im_shape, mode="xyxy")
boxlist.add_field("objectness", score)
boxlist = boxlist.clip_to_image(remove_empty=False)
boxlist = remove_small_boxes(boxlist, self.min_size)
boxlist = boxlist_nms(
boxlist,
self.nms_thresh,
max_proposals=self.post_nms_top_n,
score_field="objectness",
)
result.append(boxlist)
return result
def boxlist_box_voting(top_boxlist,
all_boxlist,
thresh,
scoring_method='ID',
beta=1.0,
score_field="scores"):
if thresh <= 0:
return top_boxlist
mode = top_boxlist.mode
top_boxes = top_boxlist.convert("xyxy").bbox
all_boxes = all_boxlist.convert("xyxy").bbox
top_score = top_boxlist.get_field(score_field)
all_score = all_boxlist.get_field(score_field)
boxes, scores = _box_voting(top_boxes, top_score, all_boxes, all_score,
thresh, scoring_method, beta)
boxlist = BoxList(boxes, all_boxlist.size, mode="xyxy")
boxlist.add_field("scores", scores)
return boxlist.convert(mode)
def prepare_boxlist(self, boxes, scores, image_shape):
"""
Returns BoxList from `boxes` and adds probability scores information
as an extra field
`boxes` has shape (#detections, 4 * #classes), where each row represents
a list of predicted bounding boxes for each of the object classes in the
dataset (including the background class). The detections in each row
originate from the same object proposal.
`scores` has shape (#detection, #classes), where each row represents a list
of object detection confidence scores for each of the object classes in the
dataset (including the background class). `scores[i, j]`` corresponds to the
box at `boxes[i, j * 4:(j + 1) * 4]`.
"""
boxes = boxes.reshape(-1, 4)
scores = scores.reshape(-1)
boxlist = BoxList(boxes, image_shape, mode="xyxy")
boxlist.add_field("scores", scores)
return boxlist
def filter_results(self, boxlist, num_classes):
"""Returns bounding-box detection results by thresholding on scores and
applying non-maximum suppression (NMS).
"""
# unwrap the boxlist to avoid additional overhead.
# if we had multi-class NMS, we could perform this directly on the boxlist
boxes = boxlist.bbox.reshape(-1, num_classes * 4)
scores = boxlist.get_field("scores").reshape(-1, num_classes)
device = scores.device
result = []
# Apply threshold on detection probabilities and apply NMS
# Skip j = 0, because it's the background class
inds_all = scores > self.score_thresh
for j in range(1, num_classes):
inds = inds_all[:, j].nonzero().squeeze(1)
scores_j = scores[inds, j]
boxes_j = boxes[inds, j * 4 : (j + 1) * 4]
boxlist_for_class = BoxList(boxes_j, boxlist.size, mode="xyxy")
boxlist_for_class.add_field("scores", scores_j)
boxlist_for_class_old = boxlist_for_class
if cfg.TEST.SOFT_NMS.ENABLED:
boxlist_for_class = boxlist_soft_nms(
boxlist_for_class,
sigma=cfg.TEST.SOFT_NMS.SIGMA,
overlap_thresh=self.nms,
score_thresh=0.0001,
method=cfg.TEST.SOFT_NMS.METHOD
)
else:
boxlist_for_class = boxlist_nms(
boxlist_for_class, self.nms
)
# Refine the post-NMS boxes using bounding-box voting
if cfg.TEST.BBOX_VOTE.ENABLED and boxes_j.shape[0] > 0:
boxlist_for_class = boxlist_box_voting(
boxlist_for_class,
boxlist_for_class_old,
cfg.TEST.BBOX_VOTE.VOTE_TH,
scoring_method=cfg.TEST.BBOX_VOTE.SCORING_METHOD
)
num_labels = len(boxlist_for_class)
boxlist_for_class.add_field(
"labels", torch.full((num_labels,), j, dtype=torch.int64, device=device)
)
result.append(boxlist_for_class)
result = cat_boxlist(result)
number_of_detections = len(result)
# Limit to max_per_image detections **over all classes**
if number_of_detections > self.detections_per_img > 0:
cls_scores = result.get_field("scores")
image_thresh, _ = torch.kthvalue(
cls_scores.cpu(), number_of_detections - self.detections_per_img + 1
)
keep = cls_scores >= image_thresh.item()
keep = torch.nonzero(keep).squeeze(1)
result = result[keep]
return result
def __getitem__(self, idx):
img, anno = super(COCODataset, self).__getitem__(idx)
# filter crowd annotations
# TODO might be better to add an extra field
if 'iscrowd' in anno[0]:
anno = [obj for obj in anno if obj["iscrowd"] == 0]
boxes = [obj["bbox"] for obj in anno]
boxes = torch.as_tensor(boxes).reshape(-1, 4) # guard against no boxes
target = BoxList(boxes, img.size, mode="xywh").convert("xyxy")
classes = [obj["category_id"] for obj in anno]
classes = [self.json_category_id_to_contiguous_id[c] for c in classes]
classes = torch.tensor(classes)
target.add_field("labels", classes)
target = target.clip_to_image(remove_empty=True)
if self._transforms is not None:
img, target = self._transforms(img, target)
return img, target, idx
def forward(self, image_list, feature_maps):
grid_sizes = [feature_map.shape[-2:] for feature_map in feature_maps]
anchors_over_all_feature_maps = self.grid_anchors(grid_sizes)
anchors = []
for i, (image_height,
image_width) in enumerate(image_list.image_sizes):
anchors_in_image = []
for anchors_per_feature_map in anchors_over_all_feature_maps:
boxlist = BoxList(anchors_per_feature_map,
(image_width, image_height),
mode="xyxy")
self.add_visibility_to(boxlist)
anchors_in_image.append(boxlist)
anchors.append(anchors_in_image)
return anchors
def parse_anns(self, image_size, anns):
boxes = [ann['bbox'] for ann in anns]
boxes = torch.as_tensor(boxes).reshape(-1, 4) # guard against no boxes
bbox = BoxList(boxes, image_size, mode="xywh").convert("xyxy")
classes = [ann['category_id'] for ann in anns]
classes = torch.tensor(classes)
bbox.add_field("labels", classes)
masks = [ann['segmentation'] for ann in anns]
masks = Mask(masks, image_size, mode='poly')
bbox.add_field("masks", masks)
parsing = [ann['parsing'] for ann in anns]
parsing = ParsingPoly(parsing, image_size)
bbox.add_field("parsing", parsing)
return bbox
def evaluate_box_proposals(predictions,
dataset,
thresholds=None,
area="all",
limit=None):
"""Evaluate detection proposal recall metrics. This function is a much
faster alternative to the official COCO API recall evaluation code. However,
it produces slightly different results.
"""
# Record max overlap value for each gt box
# Return vector of overlap values
areas = {
"all": 0,
"small": 1,
"medium": 2,
"large": 3,
"96-128": 4,
"128-256": 5,
"256-512": 6,
"512-inf": 7,
}
area_ranges = [
[0**2, 1e5**2], # all
[0**2, 32**2], # small
[32**2, 96**2], # medium
[96**2, 1e5**2], # large
[96**2, 128**2], # 96-128
[128**2, 256**2], # 128-256
[256**2, 512**2], # 256-512
[512**2, 1e5**2],
] # 512-inf
assert area in areas, "Unknown area range: {}".format(area)
area_range = area_ranges[areas[area]]
gt_overlaps = []
num_pos = 0
for image_id, prediction in enumerate(predictions):
original_id = dataset.id_to_img_map[image_id]
img_info = dataset.get_img_info(image_id)
image_width = img_info["width"]
image_height = img_info["height"]
prediction = prediction.resize((image_width, image_height))
# sort predictions in descending order
# TODO maybe remove this and make it explicit in the documentation
inds = prediction.get_field("objectness").sort(descending=True)[1]
prediction = prediction[inds]
ann_ids = dataset.coco.getAnnIds(imgIds=original_id)
anno = dataset.coco.loadAnns(ann_ids)
gt_boxes = [obj["bbox"] for obj in anno if obj["iscrowd"] == 0]
gt_boxes = torch.as_tensor(gt_boxes).reshape(
-1, 4) # guard against no boxes
gt_boxes = BoxList(gt_boxes, (image_width, image_height),
mode="xywh").convert("xyxy")
gt_areas = torch.as_tensor(
[obj["area"] for obj in anno if obj["iscrowd"] == 0])
if len(gt_boxes) == 0:
continue
valid_gt_inds = (gt_areas >= area_range[0]) & (gt_areas <=
area_range[1])
gt_boxes = gt_boxes[valid_gt_inds]
num_pos += len(gt_boxes)
if len(gt_boxes) == 0:
continue
if len(prediction) == 0:
continue
if limit is not None and len(prediction) > limit:
prediction = prediction[:limit]
overlaps = boxlist_iou(prediction, gt_boxes)
_gt_overlaps = torch.zeros(len(gt_boxes))
for j in range(min(len(prediction), len(gt_boxes))):
# find which proposal box maximally covers each gt box
# and get the iou amount of coverage for each gt box
max_overlaps, argmax_overlaps = overlaps.max(dim=0)
# find which gt box is 'best' covered (i.e. 'best' = most iou)
gt_ovr, gt_ind = max_overlaps.max(dim=0)
assert gt_ovr >= 0
# find the proposal box that covers the best covered gt box
box_ind = argmax_overlaps[gt_ind]
# record the iou coverage of this gt box
_gt_overlaps[j] = overlaps[box_ind, gt_ind]
assert _gt_overlaps[j] == gt_ovr
# mark the proposal box and the gt box as used
overlaps[box_ind, :] = -1
overlaps[:, gt_ind] = -1
# append recorded iou coverage level
gt_overlaps.append(_gt_overlaps)
gt_overlaps = torch.cat(gt_overlaps, dim=0)
gt_overlaps, _ = torch.sort(gt_overlaps)
if thresholds is None:
step = 0.05
thresholds = torch.arange(0.5, 0.95 + 1e-5, step, dtype=torch.float32)
recalls = torch.zeros_like(thresholds)
# compute recall for each iou threshold
for i, t in enumerate(thresholds):
recalls[i] = (gt_overlaps >= t).float().sum() / float(num_pos)
# ar = 2 * np.trapz(recalls, thresholds)
ar = recalls.mean()
return {
"ar": ar,
"recalls": recalls,
"thresholds": thresholds,
"gt_overlaps": gt_overlaps,
"num_pos": num_pos,
}
def forward(self, box_cls_all, box_reg_all, centerness_all, boxes_all):
device = box_cls_all.device
boxes_per_image = [len(box) for box in boxes_all]
cls = box_cls_all.split(boxes_per_image, dim=0)
reg = box_reg_all.split(boxes_per_image, dim=0)
center = centerness_all.split(boxes_per_image, dim=0)
results = []
for box_cls, box_regression, centerness, boxes in zip(cls, reg, center, boxes_all):
N, C, H, W = box_cls.shape
# put in the same format as locations
box_cls = box_cls.permute(0, 2, 3, 1).reshape(N, -1, self.num_classes).sigmoid()
box_regression = box_regression.permute(0, 2, 3, 1).reshape(N, -1, 4)
centerness = centerness.permute(0, 2, 3, 1).reshape(N, -1).sigmoid()
# multiply the classification scores with centerness scores
box_cls = box_cls * centerness[:, :, None]
_boxes = boxes.bbox
size = boxes.size
boxes_scores = boxes.get_field("scores")
results_per_image = [boxes]
for i in range(N):
box = _boxes[i]
boxes_score = boxes_scores[i]
per_box_cls = box_cls[i]
per_box_cls_max, per_box_cls_inds = per_box_cls.max(dim=0)
per_class = torch.range(2, 1 + self.num_classes, dtype=torch.long, device=device)
per_box_regression = box_regression[i]
per_box_regression = per_box_regression[per_box_cls_inds]
x_step = 1.0
y_step = 1.0
shifts_x = torch.arange(
0, self.m, step=x_step,
dtype=torch.float32, device=device
) + x_step / 2
shifts_y = torch.arange(
0, self.m, step=y_step,
dtype=torch.float32, device=device
) + y_step / 2
shift_y, shift_x = torch.meshgrid(shifts_y, shifts_x)
shift_x = shift_x.reshape(-1)
shift_y = shift_y.reshape(-1)
locations = torch.stack((shift_x, shift_y), dim=1)
per_locations = locations[per_box_cls_inds]
_x1 = per_locations[:, 0] - per_box_regression[:, 0]
_y1 = per_locations[:, 1] - per_box_regression[:, 1]
_x2 = per_locations[:, 0] + per_box_regression[:, 2]
_y2 = per_locations[:, 1] + per_box_regression[:, 3]
_x1 = _x1 / self.m * (box[2] - box[0]) + box[0]
_y1 = _y1 / self.m * (box[3] - box[1]) + box[1]
_x2 = _x2 / self.m * (box[2] - box[0]) + box[0]
_y2 = _y2 / self.m * (box[3] - box[1]) + box[1]
detections = torch.stack([_x1, _y1, _x2, _y2], dim=-1)
boxlist = BoxList(detections, size, mode="xyxy")
boxlist.add_field("labels", per_class)
boxlist.add_field("scores", torch.sqrt(torch.sqrt(per_box_cls_max) * boxes_score))
boxlist = boxlist.clip_to_image(remove_empty=False)
boxlist = remove_small_boxes(boxlist, 0)
results_per_image.append(boxlist)
results_per_image = cat_boxlist(results_per_image)
results.append(results_per_image)
return results
def filter_results(boxlist):
num_classes = cfg.MODEL.NUM_CLASSES
if not cfg.TEST.SOFT_NMS.ENABLED and not cfg.TEST.BBOX_VOTE.ENABLED:
# multiclass nms
scores = boxlist.get_field("scores")
device = scores.device
num_repeat = int(boxlist.bbox.shape[0] / num_classes)
labels = np.tile(np.arange(num_classes), num_repeat)
boxlist.add_field(
"labels",
torch.from_numpy(labels).to(dtype=torch.int64, device=device))
fg_labels = torch.from_numpy(
(np.arange(boxlist.bbox.shape[0]) % num_classes !=
0).astype(int)).to(dtype=torch.bool, device=device)
_scores = scores > cfg.FAST_RCNN.SCORE_THRESH
inds_all = _scores & fg_labels
result = boxlist_ml_nms(boxlist[inds_all], cfg.FAST_RCNN.NMS)
else:
# boxes = boxlist.bbox.reshape(-1, num_classes * 4)
boxes = boxlist.bbox.reshape(-1, 4)
labels = boxlist.get_field('labels')
scores = boxlist.get_field("scores")
# scores = boxlist.get_field("scores").reshape(-1, num_classes)
device = scores.device
result = []
# Apply threshold on detection probabilities and apply NMS
# Skip j = 0, because it's the background class
inds_all = scores > cfg.FAST_RCNN.SCORE_THRESH
for j in range(1, num_classes):
# inds = inds_all[:, j].nonzero().squeeze(1)
class_inds = labels == j
inds = (inds_all + class_inds == 2)
scores_j = scores[inds]
boxes_j = boxes[inds]
# scores_j = scores[inds, j]
# boxes_j = boxes[inds, j * 4 : (j + 1) * 4]
boxlist_for_class = BoxList(boxes_j, boxlist.size, mode="xyxy")
boxlist_for_class.add_field("scores", scores_j)
boxlist_for_class_old = boxlist_for_class
if cfg.TEST.SOFT_NMS.ENABLED:
boxlist_for_class = boxlist_soft_nms(
boxlist_for_class,
sigma=cfg.TEST.SOFT_NMS.SIGMA,
overlap_thresh=cfg.FAST_RCNN.NMS,
score_thresh=0.0001,
method=cfg.TEST.SOFT_NMS.METHOD)
else:
boxlist_for_class = boxlist_nms(boxlist_for_class,
cfg.FAST_RCNN.NMS)
# Refine the post-NMS boxes using bounding-box voting
if cfg.TEST.BBOX_VOTE.ENABLED and boxes_j.shape[0] > 0:
boxlist_for_class = boxlist_box_voting(
boxlist_for_class,
boxlist_for_class_old,
cfg.TEST.BBOX_VOTE.VOTE_TH,
scoring_method=cfg.TEST.BBOX_VOTE.SCORING_METHOD)
num_labels = len(boxlist_for_class)
boxlist_for_class.add_field(
"labels",
torch.full((num_labels, ), j, dtype=torch.int64,
device=device))
result.append(boxlist_for_class)
result = cat_boxlist(result)
number_of_detections = len(result)
# Limit to max_per_image detections **over all classes**
if number_of_detections > cfg.FAST_RCNN.DETECTIONS_PER_IMG > 0:
cls_scores = result.get_field("scores")
image_thresh, _ = torch.kthvalue(
cls_scores.cpu(),
number_of_detections - cfg.FAST_RCNN.DETECTIONS_PER_IMG + 1)
keep = cls_scores >= image_thresh.item()
keep = torch.nonzero(keep).squeeze(1)
result = result[keep]
return result
def prepare_boxlist(self, boxes, scores, image_shape):
boxes = boxes.reshape(-1, 4)
scores = scores.reshape(-1)
boxlist = BoxList(boxes, image_shape, mode="xyxy")
boxlist.add_field("scores", scores)
return boxlist