def call(self, inputs):
rois = inputs[0]
mrcnn_class = inputs[1]
mrcnn_bbox = inputs[2]
image_meta = inputs[3]
# Get windows of images in normalized coordinates. Windows are the area
# in the image that excludes the padding.
# Use the shape of the first image in the batch to normalize the window
# because we know that all images get resized to the same size.
m = utils.parse_image_meta_graph(image_meta)
image_shape = m['image_shape'][0]
window = utils.norm_boxes_graph(m['window'], image_shape[:2])
# Run detection refinement graph on each item in the batch
detections_batch = utils.batch_slice(
[rois, mrcnn_class, mrcnn_bbox, window],
lambda x, y, w, z: refine_detections_graph(x, y, w, z,
self.bbox_std_dev,
self.detection_min_confidence,
self.detection_max_instance,
self.detection_nms_threshold),
self.count_image_per_gpu)
# Reshape output
# [batch, num_detections, (y1, x1, y2, x2, class_id, class_score)] in
# normalized coordinates
return tf.reshape(
detections_batch,
[self.size_batch, self.detection_max_instance, 6])
def call(self, inputs):
rois = inputs[0]
mrcnn_class = inputs[1]
mrcnn_bbox = inputs[2]
image_meta = inputs[3]
# Get windows of images in normalized coordinates. Windows are the area
# in the image that excludes the padding.
# Use the shape of the first image in the batch to normalize the window
# because we know that all images get resized to the same size.
m = parse_image_meta_graph(image_meta)
image_shape = m['image_shape'][0]
window = norm_boxes_graph(m['window'], image_shape[:2])
# Run detection refinement graph on each item in the batch
detections_batch = utils.batch_slice([
rois, mrcnn_class, mrcnn_bbox, window
], lambda x, y, w, z: refine_detections_graph(x, y, w, z, self.config),
self.config.IMAGES_PER_GPU)
# Reshape output
# [batch, num_detections, (y1, x1, y2, x2, class_id, class_score)] in
# normalized coordinates
return tf.reshape(
detections_batch,
[self.config.BATCH_SIZE, self.config.DETECTION_MAX_INSTANCES, 6])
def call(self, inputs):
boxes = inputs[0]
image_meta = inputs[1]
feature_maps = inputs[2:]
y1, x1, y2, x2 = tf.split(boxes, 4, axis=2)
h = y2 - y1
w = x2 - x1
image_shape = parse_image_meta_graph(image_meta)['image_shape'][0]
image_area = tf.cast(image_shape[0] * image_shape[1], tf.float32)
roi_level = log2_graph(tf.sqrt(h*w) / (224.0 / tf.sqrt(image_area)))
roi_level = tf.minimum(5, tf.maximum(2, 4 + tf.cast(tf.round(roi_level), tf.int32)))
roi_level = tf.squeeze(roi_level, 2)
pooled = []
box_to_level = []
for i, level in enumerate(range(2, 6)):
ix = tf.where(tf.equal(roi_level, level))
level_boxes = tf.gather_nd(boxes, ix)
box_indices = tf.cast(ix[:, 0], tf.int32)
box_to_level.append(ix)
level_boxes = tf.stop_gradient(level_boxes)
box_indices = tf.stop_gradient(box_indices)
pooled.append(
tf.image.crop_and_resize(
feature_maps[i],
level_boxes, box_indices, self.pool_shape, method="bilinear"))
pooled = tf.concat(pooled, axis=0)
box_to_level = tf.concat(box_to_level, axis=0)
box_range = tf.expand_dims(tf.range(tf.shape(box_to_level)[0]), 1)
box_to_level = tf.concat([tf.cast(box_to_level, tf.int32), box_range], axis=1)
sorting_tensor = box_to_level[:, 0] * 100000 + box_to_level[:, 1]
ix = tf.nn.top_k(sorting_tensor, k=tf.shape(box_to_level)[0]).indices[::-1]
pooled = tf.gather(pooled, ix)
shape = tf.concat([tf.shape(boxes)[:2], tf.shape(pooled)[1:]], axis=0)
pooled = tf.reshape(pooled, shape)
return pooled
def __call__(self, ipt):
rois = ipt[0]
mrcnn_class = ipt[1]
mrcnn_bbox = ipt[2]
image_meta = ipt[3]
m = utils.parse_image_meta_graph(image_meta)
image_shape = m['image_shape'][0]
window = utils.norm_boxes(m['window'], image_shape[:2])
detections_batch = utils.batch_slice(
[rois, mrcnn_class, mrcnn_bbox, window],
lambda w, x, y, z: layer.refine_detections(w, x, y, z),
self.image_per_gpu)
return tf.reshape(
detections_batch,
[self.image_per_gpu, self.detection_max_instances, 6])
def call(self, inputs):
# Crop boxes [batch, num_boxes, (y1, x1, y2, x2)] in normalized coords
boxes = inputs[0]
# Image meta
# Holds details about the image. See compose_image_meta()
image_meta = inputs[1]
# Feature Maps. List of feature maps from different level of the
# feature pyramid. Each is [batch, height, width, channels]
feature_maps = inputs[2:]
# Assign each ROI to a level in the pyramid based on the ROI area.
y1, x1, y2, x2 = tf.split(boxes, 4, axis=2)
h = y2 - y1
w = x2 - x1
# Use shape of first image. Images in a batch must have the same size.
image_shape = utils.parse_image_meta_graph(image_meta)['image_shape'][0]
# Equation 1 in the Feature Pyramid Networks paper. Account for
# the fact that our coordinates are normalized here.
# e.g. a 224x224 ROI (in pixels) maps to P4
image_area = tf.cast(image_shape[0] * image_shape[1], tf.float32)
roi_level = self.log2_graph(tf.sqrt(h * w) / (224.0 / tf.sqrt(image_area)))
roi_level = tf.minimum(5, tf.maximum(
2, 4 + tf.cast(tf.round(roi_level), tf.int32)))
roi_level = tf.squeeze(roi_level, 2)
# Loop through levels and apply ROI pooling to each. P2 to P5.
pooled = []
box_to_level = []
for i, level in enumerate(range(2, 6)):
ix = tf.where(tf.equal(roi_level, level))
level_boxes = tf.gather_nd(boxes, ix)
# Box indices for crop_and_resize.
box_indices = tf.cast(ix[:, 0], tf.int32)
# Keep track of which box is mapped to which level
box_to_level.append(ix)
# Stop gradient propogation to ROI proposals
level_boxes = tf.stop_gradient(level_boxes)
box_indices = tf.stop_gradient(box_indices)
# Crop and Resize
# From Mask R-CNN paper: "We sample four regular locations, so
# that we can evaluate either max or average pooling. In fact,
# interpolating only a single value at each bin center (without
# pooling) is nearly as effective."
#
# Here we use the simplified approach of a single value per bin,
# which is how it's done in tf.crop_and_resize()
# Result: [batch * num_boxes, pool_height, pool_width, channels]
pooled.append(tf.image.crop_and_resize(
feature_maps[i], level_boxes, box_indices, self.pool_shape,
method="bilinear"))
# Pack pooled features into one tensor
pooled = tf.concat(pooled, axis=0)
# Pack box_to_level mapping into one array and add another
# column representing the order of pooled boxes
box_to_level = tf.concat(box_to_level, axis=0)
box_range = tf.expand_dims(tf.range(tf.shape(box_to_level)[0]), 1)
box_to_level = tf.concat([tf.cast(box_to_level, tf.int32), box_range],
axis=1)
# Rearrange pooled features to match the order of the original boxes
# Sort box_to_level by batch then box index
# TF doesn't have a way to sort by two columns, so merge them and sort.
sorting_tensor = box_to_level[:, 0] * 100000 + box_to_level[:, 1]
ix = tf.nn.top_k(sorting_tensor, k=tf.shape(
box_to_level)[0]).indices[::-1]
ix = tf.gather(box_to_level[:, 2], ix)
pooled = tf.gather(pooled, ix)
# Re-add the batch dimension
shape = tf.concat([tf.shape(boxes)[:2], tf.shape(pooled)[1:]], axis=0)
pooled = tf.reshape(pooled, shape)
return pooled
def build(self, mode, config):
"""Build Mask R-CNN architecture.
input_shape: The shape of the input image.
mode: Either "training" or "inference". The inputs and
outputs of the model differ accordingly.
"""
assert mode in ['training', 'inference']
# Image size must be dividable by 2 multiple times
h, w = config.IMAGE_SHAPE[:2]
if h / 2**6 != int(h / 2**6) or w / 2**6 != int(w / 2**6):
raise Exception(
"Image size must be dividable by 2 at least 6 times "
"to avoid fractions when downscaling and upscaling."
"For example, use 256, 320, 384, 448, 512, ... etc. ")
# Inputs
input_image = Input(shape=[None, None, config.IMAGE_SHAPE[2]],
name="input_image")
input_image_meta = Input(shape=[config.IMAGE_META_SIZE],
name="input_image_meta")
if mode == "training":
# RPN GT
input_rpn_match = Input(shape=[None, 1],
name="input_rpn_match",
dtype=tf.int32)
input_rpn_bbox = Input(shape=[None, 4],
name="input_rpn_bbox",
dtype=tf.float32)
# Detection GT (class IDs, bounding boxes, and masks)
# 1. GT Class IDs (zero padded)
input_gt_class_ids = Input(shape=[None],
name="input_gt_class_ids",
dtype=tf.int32)
# 2. GT Boxes in pixels (zero padded)
# [batch, MAX_GT_INSTANCES, (y1, x1, y2, x2)] in image coordinates
input_gt_boxes = Input(shape=[None, 4],
name="input_gt_boxes",
dtype=tf.float32)
# Normalize coordinates
gt_boxes = Lambda(lambda x: norm_boxes_graph(
x,
K.shape(input_image)[1:3]))(input_gt_boxes)
# 3. GT Masks (zero padded)
# [batch, height, width, MAX_GT_INSTANCES]
if config.USE_MINI_MASK:
input_gt_masks = Input(shape=[
config.MINI_MASK_SHAPE[0], config.MINI_MASK_SHAPE[1], None
],
name="input_gt_masks",
dtype=bool)
else:
input_gt_masks = Input(
shape=[config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1], None],
name="input_gt_masks",
dtype=bool)
elif mode == "inference":
# Anchors in normalized coordinates
input_anchors = Input(shape=[None, 4], name="input_anchors")
# Build the shared convolutional layers.
# Bottom-up Layers
# Returns a list of the last layers of each stage, 5 in total.
# Don't create the thead (stage 5), so we pick the 4th item in the list.
if callable(config.BACKBONE):
_, C2, C3, C4, C5 = config.BACKBONE(input_image,
stage5=True,
train_bn=config.TRAIN_BN)
else:
_, C2, C3, C4, C5 = resnet_graph(input_image,
config.BACKBONE,
stage5=True,
train_bn=config.TRAIN_BN)
# Top-down Layers
# TODO: add assert to varify feature map sizes match what's in configs
P5 = Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (1, 1), name='fpn_c5p5')(C5)
P4 = add([
UpSampling2D(size=(2, 2), name="fpn_p5upsampled")(P5),
Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (1, 1), name='fpn_c4p4')(C4)
])
P3 = add([
UpSampling2D(size=(2, 2), name="fpn_p4upsampled")(P4),
Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (1, 1), name='fpn_c3p3')(C3)
])
P2 = add([
UpSampling2D(size=(2, 2), name="fpn_p3upsampled")(P3),
Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (1, 1), name='fpn_c2p2')(C2)
])
# Attach 3x3 conv to all P layers to get the final feature maps.
P2 = Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (3, 3),
padding="SAME",
name="fpn_p2")(P2)
P3 = Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (3, 3),
padding="SAME",
name="fpn_p3")(P3)
P4 = Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (3, 3),
padding="SAME",
name="fpn_p4")(P4)
P5 = Conv2D(config.TOP_DOWN_PYRAMID_SIZE, (3, 3),
padding="SAME",
name="fpn_p5")(P5)
# P6 is used for the 5th anchor scale in RPN. Generated by
# subsampling from P5 with stride of 2.
P6 = MaxPooling2D(pool_size=(1, 1), strides=2, name="fpn_p6")(P5)
# Note that P6 is used in RPN, but not in the classifier heads.
rpn_feature_maps = [P2, P3, P4, P5, P6]
mrcnn_feature_maps = [P2, P3, P4, P5]
# Anchors
if mode == "training":
anchors = self.get_anchors(config.IMAGE_SHAPE)
# Duplicate across the batch dimension because Keras requires it
# TODO: can this be optimized to avoid duplicating the anchors?
anchors = np.broadcast_to(anchors,
(config.BATCH_SIZE, ) + anchors.shape)
# A hack to get around Keras's bad support for constants
anchors = Lambda(lambda x: tf.Variable(anchors),
name="anchors")(input_image)
else:
anchors = input_anchors
# RPN Model
rpn = build_rpn_model(config.RPN_ANCHOR_STRIDE,
len(config.RPN_ANCHOR_RATIOS),
config.TOP_DOWN_PYRAMID_SIZE)
# Loop through pyramid layers
layer_outputs = [] # list of lists
for p in rpn_feature_maps:
layer_outputs.append(rpn([p]))
# Concatenate layer outputs
# Convert from list of lists of level outputs to list of lists
# of outputs across levels.
# e.g. [[a1, b1, c1], [a2, b2, c2]] => [[a1, a2], [b1, b2], [c1, c2]]
output_names = ["rpn_class_logits", "rpn_class", "rpn_bbox"]
outputs = list(zip(*layer_outputs))
outputs = [
Concatenate(axis=1, name=n)(list(o))
for o, n in zip(outputs, output_names)
]
rpn_class_logits, rpn_class, rpn_bbox = outputs
# Generate proposals
# Proposals are [batch, N, (y1, x1, y2, x2)] in normalized coordinates
# and zero padded.
proposal_count = config.POST_NMS_ROIS_TRAINING if mode == "training" \
else config.POST_NMS_ROIS_INFERENCE
rpn_rois = ProposalLayer(proposal_count=proposal_count,
nms_threshold=config.RPN_NMS_THRESHOLD,
name="ROI",
config=config)([rpn_class, rpn_bbox, anchors])
if mode == "training":
# Class ID mask to mark class IDs supported by the dataset the image
# came from.
active_class_ids = Lambda(lambda x: parse_image_meta_graph(x)[
"active_class_ids"])(input_image_meta)
if not config.USE_RPN_ROIS:
# Ignore predicted ROIs and use ROIs provided as an input.
input_rois = Input(shape=[config.POST_NMS_ROIS_TRAINING, 4],
name="input_roi",
dtype=np.int32)
# Normalize coordinates
target_rois = Lambda(lambda x: norm_boxes_graph(
x,
K.shape(input_image)[1:3]))(input_rois)
else:
target_rois = rpn_rois
# Generate detection targets
# Subsamples proposals and generates target outputs for training
# Note that proposal class IDs, gt_boxes, and gt_masks are zero
# padded. Equally, returned rois and targets are zero padded.
rois, target_class_ids, target_bbox, target_mask = \
DetectionTargetLayer(config, name="proposal_targets")([
target_rois, input_gt_class_ids, gt_boxes, input_gt_masks])
# Network Heads
# TODO: verify that this handles zero padded ROIs
mrcnn_class_logits, mrcnn_class, mrcnn_bbox = \
fpn_classifier_graph(rois, mrcnn_feature_maps, input_image_meta,
config.POOL_SIZE, config.NUM_CLASSES,
train_bn=config.TRAIN_BN,
fc_layers_size=config.FPN_CLASSIF_FC_LAYERS_SIZE)
mrcnn_mask = build_fpn_mask_graph(rois,
mrcnn_feature_maps,
input_image_meta,
config.MASK_POOL_SIZE,
config.NUM_CLASSES,
train_bn=config.TRAIN_BN)
# TODO: clean up (use tf.identify if necessary)
output_rois = Lambda(lambda x: x * 1, name="output_rois")(rois)
# Losses
rpn_class_loss = Lambda(lambda x: rpn_class_loss_graph(*x),
name="rpn_class_loss")(
[input_rpn_match, rpn_class_logits])
rpn_bbox_loss = Lambda(lambda x: rpn_bbox_loss_graph(config, *x),
name="rpn_bbox_loss")([
input_rpn_bbox, input_rpn_match,
rpn_bbox
])
class_loss = Lambda(lambda x: mrcnn_class_loss_graph(*x),
name="mrcnn_class_loss")([
target_class_ids, mrcnn_class_logits,
active_class_ids
])
bbox_loss = Lambda(lambda x: mrcnn_bbox_loss_graph(*x),
name="mrcnn_bbox_loss")(
[target_bbox, target_class_ids, mrcnn_bbox])
mask_loss = Lambda(lambda x: mrcnn_mask_loss_graph(*x),
name="mrcnn_mask_loss")(
[target_mask, target_class_ids, mrcnn_mask])
# Model
inputs = [
input_image, input_image_meta, input_rpn_match, input_rpn_bbox,
input_gt_class_ids, input_gt_boxes, input_gt_masks
]
if not config.USE_RPN_ROIS:
inputs.append(input_rois)
outputs = [
rpn_class_logits, rpn_class, rpn_bbox, mrcnn_class_logits,
mrcnn_class, mrcnn_bbox, mrcnn_mask, rpn_rois, output_rois,
rpn_class_loss, rpn_bbox_loss, class_loss, bbox_loss, mask_loss
]
model = Model(inputs, outputs, name='mask_rcnn')
else:
# Network Heads
# Proposal classifier and BBox regressor heads
mrcnn_class_logits, mrcnn_class, mrcnn_bbox = \
fpn_classifier_graph(rpn_rois, mrcnn_feature_maps, input_image_meta,
config.POOL_SIZE, config.NUM_CLASSES,
train_bn=config.TRAIN_BN,
fc_layers_size=config.FPN_CLASSIF_FC_LAYERS_SIZE)
# Detections
# output is [batch, num_detections, (y1, x1, y2, x2, class_id, score)] in
# normalized coordinates
detections = DetectionLayer(config, name="mrcnn_detection")(
[rpn_rois, mrcnn_class, mrcnn_bbox, input_image_meta])
# Create masks for detections
detection_boxes = Lambda(lambda x: x[..., :4])(detections)
mrcnn_mask = build_fpn_mask_graph(detection_boxes,
mrcnn_feature_maps,
input_image_meta,
config.MASK_POOL_SIZE,
config.NUM_CLASSES,
train_bn=config.TRAIN_BN)
model = Model([input_image, input_image_meta, input_anchors], [
detections, mrcnn_class, mrcnn_bbox, mrcnn_mask, rpn_rois,
rpn_class, rpn_bbox
],
name='mask_rcnn')
# Add multi-GPU support.
# if configs.GPU_COUNT > 1:
# from mrcnn.parallel_model import ParallelModel
# model = ParallelModel(model, configs.GPU_COUNT)
return model
def model(self):
h, w = self.image_shape[:2]
if h / 2**6 != int(h / 2**6) or w / 2**6 != int(w / 2**6):
raise Exception(
"Image size must be dividable by 2 at least 6 times "
"to avoid fractions when downscaling and upscaling."
"For example, use 256, 320, 384, 448, 512, ... etc. ")
if self.use_pretrained_model:
c2, c3, c4, c5 = \
self.end_points['resnet_v2_50/block1/unit_2/bottleneck_v2'], \
self.end_points['resnet_v2_50/block2/unit_3/bottleneck_v2'], \
self.end_points['resnet_v2_50/block3/unit_4/bottleneck_v2'], \
self.end_points['resnet_v2_50/block4']
else:
if callable(self.backbone):
_, c2, c3, c4, c5 = self.backbone(self.input_image,
stage5=self.stage5,
is_training=self.is_training)
else:
_, c2, c3, c4, c5 = self.resnet(self.input_image)
p2, p3, p4, p5, p6 = self.fpn([c2, c3, c4, c5])
rpn_feature_maps = [p2, p3, p4, p5, p6]
mrcnn_feature_maps = [p2, p3, p4, p5]
if self.mode == 'training':
anchors = self.get_anchors(self.image_shape)
anchors = np.broadcast_to(anchors,
(self.batch_size, ) + anchors.shape)
anchors = tf.constant(anchors)
else:
anchors = self.input_anchors
layer_outputs = []
for p in rpn_feature_maps:
layer_outputs.append(self.rpn(p))
output_names = ["rpn_class_logits", "rpn_class", "rpn_bbox"]
outputs = list(zip(*layer_outputs))
outputs = [
tf.concat(list(o), name=n, axis=1)
for o, n in zip(outputs, output_names)
]
rpn_class_logits, rpn_class, rpn_bbox = outputs
rpn_rois = self.proposal([rpn_class, rpn_bbox, anchors])
if self.mode == 'training':
active_class_ids = utils.parse_image_meta_graph(
self.input_image_meta)['active_class_ids']
if not self.use_rpn_rois:
input_rois = tf.placeholder(
dtype=tf.int32,
shape=[None, self.post_nms_rois_training, 4],
name='input_rois')
target_rois = utils.norm_boxes_graph(
input_rois,
tf.shape(self.input_image)[1:3])
else:
target_rois = rpn_rois
rois, target_class_ids, target_bbox, target_mask = \
self.targetDetection([target_rois, self.input_gt_class_ids,
self.input_gt_boxes_normalized, self.input_gt_masks])
pooled = self.pyramidRoiPooling([rois, self.input_image_meta] +
mrcnn_feature_maps)
pooled_mask = self.pyramidRoiPooling(
[rois, self.input_image_meta] + mrcnn_feature_maps,
pool_size=14)
mrcnn_class_logits, mrcnn_class, mrcnn_bbox = self.fpnClassifier(
pooled)
mrcnn_mask = self.fpnMask(pooled_mask)
output_rois = tf.identity(rois, name='output_rois')
rpn_class_loss = layer.rpn_loss(self.input_rpn_match,
rpn_class_logits)
rpn_bbox_loss = layer.rpn_bbox_loss(self.input_rpn_boxes,
self.input_rpn_match, rpn_bbox)
class_loss = layer.mrcnn_class_loss(target_class_ids,
mrcnn_class_logits,
active_class_ids)
bbox_loss = layer.mrcnn_bbox_loss(target_bbox, target_class_ids,
mrcnn_bbox)
mask_loss = layer.mrcnn_mask_loss(target_mask, target_class_ids,
mrcnn_mask)
tf.summary.scalar('rpn_class_loss', rpn_class_loss)
tf.summary.scalar('rpn_bbox_loss', rpn_bbox_loss)
tf.summary.scalar('mrcnn_class_loss', class_loss)
tf.summary.scalar('mrcnn_bbox_loss', bbox_loss)
tf.summary.scalar('mrcnn_mask_loss', mask_loss)
outputs = [
rpn_class_logits, rpn_class, rpn_bbox, mrcnn_class_logits,
mrcnn_class, mrcnn_bbox, mrcnn_mask, rpn_rois, output_rois,
rpn_class_loss, rpn_bbox_loss, class_loss, bbox_loss, mask_loss
]
else:
pooled = self.pyramidRoiPooling([rpn_rois, self.input_image_meta] +
mrcnn_feature_maps)
mrcnn_class_logits, mrcnn_class, mrcnn_bbox = self.fpnClassifier(
pooled)
detections = self.objDetection(
[rpn_rois, mrcnn_class, mrcnn_bbox, self.input_image_meta])
detections_bbox = detections[..., :4]
pooled = self.pyramidRoiPooling(
[detections_bbox, self.input_image_meta] + mrcnn_feature_maps,
pool_size=14)
mrcnn_mask = self.fpnMask(pooled)
outputs = [
detections, mrcnn_class, mrcnn_bbox, mrcnn_mask, rpn_rois,
rpn_class, rpn_bbox
]
return outputs
def __call__(self, ipt, pool_size=None):
if pool_size is not None:
self.roi_size = pool_size
boxes = ipt[0]
image_meta = ipt[1]
feature_maps = ipt[2:]
y1, x1, y2, x2 = tf.split(boxes, [1, 1, 1, 1], axis=2)
w = x2 - x1
h = y2 - y1
image_shape = utils.parse_image_meta_graph(image_meta)['image_shape'][0]
image_area = tf.cast(image_shape[0] * image_shape[1], tf.float32)
roi_level = ops.log2_graph(tf.sqrt(w * h) / (224.0 / tf.sqrt(image_area)))
roi_level = tf.minimum(5, tf.maximum(
2, 4 + tf.cast(tf.round(roi_level), tf.int32)))
roi_level = tf.squeeze(roi_level, 2)
pooled = []
box_to_level = []
for i, level in enumerate(range(2, 6)):
ix = tf.where(tf.equal(roi_level, level))
level_boxes = tf.gather_nd(boxes, ix)
# Box indices for crop_and_resize.
box_indices = tf.cast(ix[:, 0], tf.int32)
# Keep track of which box is mapped to which level
box_to_level.append(ix)
# Stop gradient propogation to ROI proposals
level_boxes = tf.stop_gradient(level_boxes)
box_indices = tf.stop_gradient(box_indices)
# Crop and Resize
# From Mask R-CNN paper: "We sample four regular locations, so
# that we can evaluate either max or average pooling. In fact,
# interpolating only a single value at each bin center (without
# pooling) is nearly as effective."
#
# Here we use the simplified approach of a single value per bin,
# which is how it's done in tf.crop_and_resize()
# Result: [batch * num_boxes, pool_height, pool_width, channels]
pooled.append(tf.image.crop_and_resize(
feature_maps[i], level_boxes, box_indices, [self.roi_size, self.roi_size],
method="bilinear"))
pooled = tf.concat(pooled, axis=0)
box_to_level = tf.concat(box_to_level, axis=0)
box_range = tf.expand_dims(tf.range(tf.shape(box_to_level)[0]), 1)
box_to_level = tf.concat([tf.cast(box_to_level, tf.int32), box_range],
axis=1)
sorting_tensor = box_to_level[:, 0] * 100000 + box_to_level[:, 1]
ix = tf.nn.top_k(sorting_tensor, k=tf.shape(
box_to_level)[0]).indices[::-1]
ix = tf.gather(box_to_level[:, 2], ix)
pooled = tf.gather(pooled, ix)
shape = tf.concat([tf.shape(boxes)[:2], tf.shape(pooled)[1:]], axis=0)
pooled = tf.reshape(pooled, shape)
return pooled