def preprocess(example, output_size=(HEIGHT, WIDTH), max_objects=50, training=True):
image, objects, image_id = decode(example)
if training:
image = random_resize(image, output_size, ratio_range=(0.5, 2.0))
image, objects = random_crop(image, objects, output_size)
image, objects = random_hflip(image, objects, 0.5)
else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939], [58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels = get(['gt_bbox', 'gt_label'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
objects = {**objects, 'gt_bbox': gt_bboxes}
bbox_targets, labels, centerness = atss_match(
gt_bboxes, gt_labels, anchors, num_level_bboxes, topk=9, centerness=True)
objects = pad_objects(objects, max_objects)
image = tf.cast(image, tf.bfloat16)
return image, {'bbox_target': bbox_targets, 'label': labels, 'centerness': centerness,
**objects, 'image_id': image_id}
def transform(example, output_size=(HEIGHT, WIDTH), training=True):
image, objects, image_id = decode(example)
# if training:
# image = random_resize(image, output_size, ratio_range=(0.8, 1.2))
# image, objects = random_crop(image, objects, output_size)
# image, objects = random_hflip(image, objects, 0.5)
# else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939],
[58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels = get(['gt_bbox', 'gt_label'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
bbox_targets, labels = atss_match(gt_bboxes,
gt_labels,
anchors,
num_level_bboxes,
topk=9,
centerness=False)
return image, {
'bbox_target': bbox_targets,
'label': labels,
'image_id': image_id
}
def preprocess(example,
output_size=(HEIGHT, WIDTH),
max_objects=50,
training=True):
image, objects, image_id = decode(example)
# if training:
# image = random_resize(image, output_size, ratio_range=(0.8, 1.2))
# image, objects = random_crop(image, objects, output_size)
# image, objects = random_hflip(image, objects, 0.5)
# else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939],
[58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels = get(['gt_bbox', 'gt_label'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
objects = {**objects, 'gt_bbox': gt_bboxes}
bbox_targets, labels, ignore = max_iou_match(gt_bboxes,
gt_labels,
bbox_coder,
pos_iou_thr=0.5,
neg_iou_thr=0.4)
objects = pad_objects(objects, max_objects)
return image, {
'bbox_target': bbox_targets,
'label': labels,
'ignore': ignore,
**objects, 'image_id': image_id
}
def preprocess(example,
output_size=(HEIGHT, WIDTH),
max_objects=50,
training=True):
image, objects, image_id = decode(example)
# if training:
# image = random_resize(image, output_size, ratio_range=(0.8, 1.2))
# image, objects = random_crop(image, objects, output_size)
# image, objects = random_hflip(image, objects, 0.5)
# else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939],
[58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels, is_difficults = get(
['bbox', 'label', 'is_difficult'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
assigned_gt_inds = atss_assign(anchors,
num_level_bboxes,
gt_bboxes,
topk=9)
bbox_targets, labels, centerness, ignore = encode_target(gt_bboxes,
gt_labels,
assigned_gt_inds,
bbox_coder,
encode_bbox=False,
centerness=True)
gt_bboxes = pad_to_fixed_size(gt_bboxes, max_objects)
gt_labels = pad_to_fixed_size(gt_labels, max_objects)
is_difficults = pad_to_fixed_size(is_difficults, max_objects)
# image = tf.cast(image, tf.bfloat16)
return image, {
'bbox_target': bbox_targets,
'label': labels,
'ignore': ignore,
'centerness': centerness,
'gt_bbox': gt_bboxes,
'gt_label': gt_labels,
'is_difficult': is_difficults,
'image_id': image_id
}
def preprocess(example,
output_size=(HEIGHT, WIDTH),
max_objects=50,
training=True):
image, objects, image_id = decode(example)
# if training:
# image = random_resize(image, output_size, ratio_range=(0.8, 1.2))
# image, objects = random_crop(image, objects, output_size)
# image, objects = random_hflip(image, objects, 0.5)
# else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939],
[58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels = get(['gt_bbox', 'gt_label'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
objects = {**objects, 'gt_bbox': gt_bboxes}
bbox_targets, labels, centerness = fcos_match(gt_bboxes,
gt_labels,
points,
num_level_points,
strides=strides,
radius=0.5)
objects = pad_objects(objects, max_objects)
return image, {
'bbox_target': bbox_targets,
'label': labels,
'centerness': centerness,
**objects, 'image_id': image_id
}
skip_prefetch=True))
it = iter(ds_train)
example = next(it)
image, objects, image_id = decode(example)
# if training:
# image = random_resize(image, output_size, ratio_range=(0.8, 1.2))
# image, objects = random_crop(image, objects, output_size)
# image, objects = random_hflip(image, objects, 0.5)
# else:
image = resize(image, output_size)
image = normalize(image, [123.68, 116.779, 103.939], [58.393, 57.12, 57.375])
image, objects = pad_to(image, objects, output_size)
gt_bboxes, gt_labels = get(['gt_bbox', 'gt_label'], objects)
gt_bboxes = coords_to_absolute(gt_bboxes, tf.shape(image)[:2])
objects = {**objects, 'gt_bbox': gt_bboxes}
bbox_targets, labels, centerness = atss_match(gt_bboxes,
gt_labels,
anchors,
num_level_bboxes,
topk=9,
centerness=True)
objects = pad_objects(objects, max_objects)