def __call__(self, data, label, gt):
#-----------------------------------------------------------------------
# Check whether to sample or not
#-----------------------------------------------------------------------
if not self.sample:
return data, label, gt
#-----------------------------------------------------------------------
# Retry sampling a couple of times
#-----------------------------------------------------------------------
source_boxes = anchors2array(gt.boxes, gt.imgsize)
box = None
box_arr = None
for _ in range(self.max_trials):
#-------------------------------------------------------------------
# Sample a bounding box
#-------------------------------------------------------------------
scale = random.uniform(self.min_scale, self.max_scale)
aspect_ratio = random.uniform(self.min_aspect_ratio,
self.max_aspect_ratio)
# make sure width and height will not be larger than 1
aspect_ratio = max(aspect_ratio, scale**2)
aspect_ratio = min(aspect_ratio, 1 / (scale**2))
width = scale * sqrt(aspect_ratio)
height = scale / sqrt(aspect_ratio)
cx = 0.5 * width + random.uniform(0, 1 - width)
cy = 0.5 * height + random.uniform(0, 1 - height)
center = Point(cx, cy)
size = Size(width, height)
#-------------------------------------------------------------------
# Check if the box satisfies the jaccard overlap constraint
#-------------------------------------------------------------------
box_arr = np.array(prop2abs(center, size, gt.imgsize))
overlap = compute_overlap(box_arr, source_boxes, 0)
if overlap.best and overlap.best.score >= self.min_jaccard_overlap:
box = Box(None, None, center, size)
break
if box is None:
return None
#-----------------------------------------------------------------------
# Crop the box and adjust the ground truth
#-----------------------------------------------------------------------
new_size = Size(box_arr[1] - box_arr[0], box_arr[3] - box_arr[2])
w_off = -box_arr[0]
h_off = -box_arr[2]
data = data[box_arr[2]:box_arr[3], box_arr[0]:box_arr[1]]
gt = transform_gt(gt, new_size, h_off, w_off)
return data, label, gt
def __call__(self, data, label, gt, seg_gt, seg_gt_to_compare):
#-----------------------------------------------------------------------
# Initialize the data vector and other variables
#-----------------------------------------------------------------------
if not self.initialized:
self.initialize()
vec = np.zeros((self.vheight, self.vwidth), dtype=np.float32)
#-----------------------------------------------------------------------
# For every box compute the best match and all the matches above 0.5
# Jaccard overlap
#-----------------------------------------------------------------------
overlaps = {}
for box in gt.boxes:
box_arr = box2array(box, self.img_size)
overlaps[box] = compute_overlap(box_arr, self.anchors_arr, 0.5)
#-----------------------------------------------------------------------
# Set up the training vector resolving conflicts in favor of a better
# match
#-----------------------------------------------------------------------
vec[:, self.num_classes] = 1 # background class
vec[:, self.num_classes + 1] = 0 # x offset
vec[:, self.num_classes + 2] = 0 # y offset
vec[:, self.num_classes + 3] = 0 # log width scale
vec[:, self.num_classes + 4] = 0 # log height scale
matches = {}
for box in gt.boxes:
for overlap in overlaps[box].good:
anchor = self.anchors[overlap.idx]
process_overlap(overlap, box, anchor, matches,
self.num_classes, vec)
matches = {}
for box in gt.boxes:
overlap = overlaps[box].best
if not overlap:
continue
anchor = self.anchors[overlap.idx]
process_overlap(overlap, box, anchor, matches, self.num_classes,
vec)
return data, vec, gt, seg_gt, seg_gt_to_compare
def __call__(self, data, label, gt):
if not self.initialized:
self.initialize()
vec = np.zeros((self.vheight, self.vwidth), dtype=np.float32)
overlaps = {}
for box in gt.boxes:
# box = 0~1
box_arr = box2array(
box, self.img_size
) # Convert proportional center-width bounds to absolute min-max bounds
# box_arr = (0~300, 0~300)
# gt_box xmin, xmax, ymin, ymax
overlaps[box] = compute_overlap(box_arr, self.anchors_arr, 0.4)
vec[:, self.num_classes] = 1. # background
vec[:, self.num_classes + 1] = 0. # x offset
vec[:, self.num_classes + 2] = 0. # y offset
vec[:, self.num_classes + 3] = 0. # log width scale
vec[:, self.num_classes + 4] = 0. # log height scale
matches = {}
for box in gt.boxes:
for overlap in overlaps[box].good:
anchor = self.anchors[overlap.idx]
process_overlap(overlap, box, anchor, matches,
self.num_classes, vec)
matches = {}
for box in gt.boxes:
overlap = overlaps[box].best
if not overlap:
continue
anchor = self.anchors[overlap.idx]
process_overlap(overlap, box, anchor, matches, self.num_classes,
vec)
return data, vec, gt.filename
def __call__(self, data, label, gt):
#-----------------------------------------------------------------------
# Check whether to sample or not
#-----------------------------------------------------------------------
if not self.sample:
return data, label, gt
#-----------------------------------------------------------------------
# Retry sampling a couple of times
#-----------------------------------------------------------------------
source_boxes = anchors2array(
gt.boxes, gt.imgsize) # get abs value(xmin,xmax,ymin,ymax)
box = None
box_arr = None
for _ in range(self.max_trials):
#-------------------------------------------------------------------
# Sample a bounding box
#-------------------------------------------------------------------
# min_scale=0.3, max_scale=1.0,
# min_aspect_ratio=0.5, max_aspect_ratio=2.0,
scale = random.uniform(self.min_scale, self.max_scale)
aspect_ratio = random.uniform(self.min_aspect_ratio,
self.max_aspect_ratio)
# make sure width and height will not be larger than 1
aspect_ratio = max(aspect_ratio, scale**2)
aspect_ratio = min(aspect_ratio, 1 / (scale**2))
width = scale * sqrt(aspect_ratio)
height = scale / sqrt(aspect_ratio)
cx = 0.5 * width + random.uniform(0, 1 - width)
cy = 0.5 * height + random.uniform(0, 1 - height)
center = Point(cx, cy)
size = Size(width, height)
#-------------------------------------------------------------------
# Check if the box satisfies the jaccard overlap constraint
#-------------------------------------------------------------------
box_arr = np.array(prop2abs(center, size, gt.imgsize))
overlap = compute_overlap(box_arr, source_boxes, 0)
if overlap.best and overlap.best.score >= self.min_jaccard_overlap:
box = Box(None, None, center, size)
break
if box is None:
return None
#-----------------------------------------------------------------------
# Crop the box and adjust the ground truth
#-----------------------------------------------------------------------
new_size = Size(box_arr[1] - box_arr[0], box_arr[3] - box_arr[2])
w_off = -box_arr[0]
h_off = -box_arr[2]
data = data.crop(
(box_arr[0], box_arr[2], box_arr[1], box_arr[3])
) # gt와 gt근처의 가상의 anchor를 잡은뒤, iou 0.1, 0.3, 0.5, 0.7, 0.9 이상이면 그 가상의 anchor부분을 crop하고 gt 좌표도 같이 변경
# 즉 일부로 gt와 iou 연관있는 그림부분을 crop함
gt = transform_gt(gt, new_size, h_off, w_off)
return data, label, gt # change gt and image_size
