def _create_cell_anchors():
"""
Generate all types of anchors for all fpn levels/scales/aspect ratios.
This function is called only once at the beginning of inference.
"""
k_max, k_min = cfg.FPN.RPN_MAX_LEVEL, cfg.FPN.RPN_MIN_LEVEL
scales_per_octave = cfg.RETINANET.SCALES_PER_OCTAVE
aspect_ratios = cfg.RETINANET.ASPECT_RATIOS
anchor_scale = cfg.RETINANET.ANCHOR_SCALE
A = scales_per_octave * len(aspect_ratios)
anchors = {}
for lvl in range(k_min, k_max + 1):
# create cell anchors array
stride = 2.**lvl
cell_anchors = np.zeros((A, 4))
a = 0
for octave in range(scales_per_octave):
octave_scale = 2**(octave / float(scales_per_octave))
for aspect in aspect_ratios:
anchor_sizes = (stride * octave_scale * anchor_scale, )
anchor_aspect_ratios = (aspect, )
cell_anchors[a, :] = generate_anchors(
stride=stride,
sizes=anchor_sizes,
aspect_ratios=anchor_aspect_ratios)
a += 1
anchors[lvl] = cell_anchors
return anchors
def get_anchors(spatial_scale, anchor_sizes):
anchors = generate_anchors.generate_anchors(
stride=1.0 / spatial_scale,
sizes=anchor_sizes,
aspect_ratios=cfg.RPN.ASPECT_RATIOS,
).astype(np.float32)
return anchors
def _create_cell_anchors():
"""
Generate all types of anchors for all fpn levels/scales/aspect ratios.
This function is called only once at the beginning of inference.
"""
k_max, k_min = cfg.FPN.RPN_MAX_LEVEL, cfg.FPN.RPN_MIN_LEVEL
scales_per_octave = cfg.RETINANET.SCALES_PER_OCTAVE
aspect_ratios = cfg.RETINANET.ASPECT_RATIOS
anchor_scale = cfg.RETINANET.ANCHOR_SCALE
A = scales_per_octave * len(aspect_ratios)
anchors = {}
for lvl in range(k_min, k_max + 1):
# create cell anchors array
stride = 2. ** lvl
cell_anchors = np.zeros((A, 4))
a = 0
for octave in range(scales_per_octave):
octave_scale = 2 ** (octave / float(scales_per_octave))
for aspect in aspect_ratios:
anchor_sizes = (stride * octave_scale * anchor_scale, )
anchor_aspect_ratios = (aspect, )
cell_anchors[a, :] = generate_anchors(
stride=stride, sizes=anchor_sizes,
aspect_ratios=anchor_aspect_ratios)
a += 1
anchors[lvl] = cell_anchors
return anchors
def add_single_scale_rpn_outputs(model, blob_in, dim_in, spatial_scale):
"""Add RPN outputs to a single scale model (i.e., no FPN)."""
anchors = generate_anchors(stride=1. / spatial_scale,
sizes=cfg.RPN.SIZES,
aspect_ratios=cfg.RPN.ASPECT_RATIOS)
num_anchors = anchors.shape[0]
dim_out = dim_in
# RPN hidden representation
model.Conv(blob_in,
'conv_rpn',
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
model.Relu('conv_rpn', 'conv_rpn')
# Proposal classification scores
model.Conv('conv_rpn',
'rpn_cls_logits',
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
# Proposal bbox regression deltas
model.Conv('conv_rpn',
'rpn_bbox_pred',
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
if not model.train or cfg.MODEL.FASTER_RCNN:
# Proposals are needed during:
# 1) inference (== not model.train) for RPN only and Faster R-CNN
# OR
# 2) training for Faster R-CNN
# Otherwise (== training for RPN only), proposals are not needed
model.net.Sigmoid('rpn_cls_logits', 'rpn_cls_probs')
model.GenerateProposals(['rpn_cls_probs', 'rpn_bbox_pred', 'im_info'],
['rpn_rois', 'rpn_roi_probs'],
anchors=anchors,
spatial_scale=spatial_scale)
if cfg.MODEL.FASTER_RCNN:
if model.train:
# Add op that generates training labels for in-network RPN proposals
model.GenerateProposalLabels(['rpn_rois', 'roidb', 'im_info'])
else:
# Alias rois to rpn_rois for inference
model.net.Alias('rpn_rois', 'rois')
# Alias da_rois to rpn_rois for inference
model.net.Alias('rpn_rois', 'da_rois')
def get_field_of_anchors(
stride, anchor_sizes, anchor_aspect_ratios, octave=None, aspect=None
):
global _threadlocal_foa
if not hasattr(_threadlocal_foa, 'cache'):
_threadlocal_foa.cache = {}
cache_key = str(stride) + str(anchor_sizes) + str(anchor_aspect_ratios)
if cache_key in _threadlocal_foa.cache:
return _threadlocal_foa.cache[cache_key]
# Anchors at a single feature cell
cell_anchors = generate_anchors(
stride=stride, sizes=anchor_sizes, aspect_ratios=anchor_aspect_ratios
)
num_cell_anchors = cell_anchors.shape[0]
# Generate canonical proposals from shifted anchors
# Enumerate all shifted positions on the (H, W) grid
fpn_max_size = cfg.FPN.COARSEST_STRIDE * np.ceil(
cfg.TRAIN.MAX_SIZE / float(cfg.FPN.COARSEST_STRIDE)
)
field_size = int(np.ceil(fpn_max_size / float(stride)))
shifts = np.arange(0, field_size) * stride
shift_x, shift_y = np.meshgrid(shifts, shifts)
shift_x = shift_x.ravel()
shift_y = shift_y.ravel()
shifts = np.vstack((shift_x, shift_y, shift_x, shift_y)).transpose()
# Broacast anchors over shifts to enumerate all anchors at all positions
# in the (H, W) grid:
# - add A cell anchors of shape (1, A, 4) to
# - K shifts of shape (K, 1, 4) to get
# - all shifted anchors of shape (K, A, 4)
# - reshape to (K*A, 4) shifted anchors
A = num_cell_anchors
K = shifts.shape[0]
field_of_anchors = (
cell_anchors.reshape((1, A, 4)) +
shifts.reshape((1, K, 4)).transpose((1, 0, 2))
)
field_of_anchors = field_of_anchors.reshape((K * A, 4))
foa = FieldOfAnchors(
field_of_anchors=field_of_anchors.astype(np.float32),
num_cell_anchors=num_cell_anchors,
stride=stride,
field_size=field_size,
octave=octave,
aspect=aspect
)
_threadlocal_foa.cache[cache_key] = foa
return foa
def get_field_of_anchors(
stride, anchor_sizes, anchor_aspect_ratios, octave=None, aspect=None
):
global _threadlocal_foa
if not hasattr(_threadlocal_foa, 'cache'):
_threadlocal_foa.cache = {}
cache_key = str(stride) + str(anchor_sizes) + str(anchor_aspect_ratios)
if cache_key in _threadlocal_foa.cache:
return _threadlocal_foa.cache[cache_key]
# Anchors at a single feature cell
cell_anchors = generate_anchors(
stride=stride, sizes=anchor_sizes, aspect_ratios=anchor_aspect_ratios
)
num_cell_anchors = cell_anchors.shape[0]
# Generate canonical proposals from shifted anchors
# Enumerate all shifted positions on the (H, W) grid
fpn_max_size = cfg.FPN.COARSEST_STRIDE * np.ceil(
cfg.TRAIN.MAX_SIZE / float(cfg.FPN.COARSEST_STRIDE)
)
field_size = int(np.ceil(fpn_max_size / float(stride)))
shifts = np.arange(0, field_size) * stride
shift_x, shift_y = np.meshgrid(shifts, shifts)
shift_x = shift_x.ravel()
shift_y = shift_y.ravel()
shifts = np.vstack((shift_x, shift_y, shift_x, shift_y)).transpose()
# Broacast anchors over shifts to enumerate all anchors at all positions
# in the (H, W) grid:
# - add A cell anchors of shape (1, A, 4) to
# - K shifts of shape (K, 1, 4) to get
# - all shifted anchors of shape (K, A, 4)
# - reshape to (K*A, 4) shifted anchors
A = num_cell_anchors
K = shifts.shape[0]
field_of_anchors = (
cell_anchors.reshape((1, A, 4)) +
shifts.reshape((1, K, 4)).transpose((1, 0, 2))
)
field_of_anchors = field_of_anchors.reshape((K * A, 4))
foa = FieldOfAnchors(
field_of_anchors=field_of_anchors.astype(np.float32),
num_cell_anchors=num_cell_anchors,
stride=stride,
field_size=field_size,
octave=octave,
aspect=aspect
)
_threadlocal_foa.cache[cache_key] = foa
return foa
def add_fpn_rpn_outputs(model, blobs_in, dim_in, spatial_scales):
"""Add RPN on FPN specific outputs."""
num_anchors = len(cfg.FPN.RPN_ASPECT_RATIOS)
dim_out = dim_in
k_max = cfg.FPN.RPN_MAX_LEVEL # coarsest level of pyramid
k_min = cfg.FPN.RPN_MIN_LEVEL # finest level of pyramid
assert len(blobs_in) == k_max - k_min + 1
for lvl in range(k_min, k_max + 1):
bl_in = blobs_in[k_max - lvl] # blobs_in is in reversed order
sc = spatial_scales[k_max - lvl] # in reversed order
slvl = str(lvl)
if lvl == k_min:
# Create conv ops with randomly initialized weights and
# zeroed biases for the first FPN level; these will be shared by
# all other FPN levels
# RPN hidden representation
conv_rpn_fpn = model.Conv(bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
rpn_cls_logits_fpn = model.Conv(conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
# Proposal bbox regression deltas
rpn_bbox_pred_fpn = model.Conv(conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
else:
# Share weights and biases
sk_min = str(k_min)
# RPN hidden representation
conv_rpn_fpn = model.ConvShared(
bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight='conv_rpn_fpn' + sk_min + '_w',
bias='conv_rpn_fpn' + sk_min + '_b')
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
rpn_cls_logits_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_cls_logits_fpn' + sk_min + '_w',
bias='rpn_cls_logits_fpn' + sk_min + '_b')
# Proposal bbox regression deltas
rpn_bbox_pred_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_bbox_pred_fpn' + sk_min + '_w',
bias='rpn_bbox_pred_fpn' + sk_min + '_b')
if not model.train or cfg.MODEL.FASTER_RCNN:
# Proposals are needed during:
# 1) inference (== not model.train) for RPN only and Faster R-CNN
# OR
# 2) training for Faster R-CNN
# Otherwise (== training for RPN only), proposals are not needed
lvl_anchors = generate_anchors(
stride=2.**lvl,
sizes=(cfg.FPN.RPN_ANCHOR_START_SIZE * 2.**(lvl - k_min), ),
aspect_ratios=cfg.FPN.RPN_ASPECT_RATIOS)
rpn_cls_probs_fpn = model.net.Sigmoid(rpn_cls_logits_fpn,
'rpn_cls_probs_fpn' + slvl)
model.GenerateProposals(
[rpn_cls_probs_fpn, rpn_bbox_pred_fpn, 'im_info'],
['rpn_rois_fpn' + slvl, 'rpn_roi_probs_fpn' + slvl],
anchors=lvl_anchors,
spatial_scale=sc)
def get_anchors(spatial_scale):
anchors = generate_anchors.generate_anchors(
stride=1. / spatial_scale,
sizes=cfg.RPN.SIZES,
aspect_ratios=cfg.RPN.ASPECT_RATIOS).astype(np.float32)
return anchors
def get_anchors(spatial_scale):
anchors = generate_anchors.generate_anchors(
stride=1. / spatial_scale,
sizes=cfg.RPN.SIZES,
aspect_ratios=cfg.RPN.ASPECT_RATIOS).astype(np.float32)
return anchors
def add_single_scale_rpn_outputs(model, blob_in, dim_in, spatial_scale):
"""Add RPN outputs to a single scale model (i.e., no FPN)."""
anchors = generate_anchors(
stride=1./ spatial_scale,
sizes=cfg.RPN.SIZES,
aspect_ratios=cfg.RPN.ASPECT_RATIOS
)
num_anchors = anchors.shape[0]
dim_out = dim_in
# RPN hidden representation
model.Conv(
blob_in,
'conv_rpn',
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
model.Relu('conv_rpn', 'conv_rpn')
# Proposal classification scores
model.Conv(
'conv_rpn',
'rpn_cls_logits',
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
#
model.Conv(
'conv_rpn',
'rpn_bbox_pred',
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
if not model.train or cfg.MODEL.FASTER_RCNN:
model.net.Sigmoid('rpn_cls_logits', 'rpn_cls_probs')
model.GenerateProposals(
['rpn_cls_probs', 'rpn_bbox_pred','im_info'],
['rpn_rois', 'rpn_roi_probs'],
anchors=anchors,
spatial_scale=spatial_scale
)
if cfg.MODEL.FASTER_RCNN:
if model.train:
model.GenerateProposalLabels(['rpn_rois', 'roidb', 'im_info'])
else:
model.net.Alias('rpn_rois', 'rois')
def add_fpn_rpn_outputs(model, blobs_in, dim_in, spatial_scales):
"""Add RPN on FPN specific outputs."""
num_anchors = len(cfg.FPN.RPN_ASPECT_RATIOS)
dim_out = dim_in
k_max = cfg.FPN.RPN_MAX_LEVEL # coarsest level of pyramid
k_min = cfg.FPN.RPN_MIN_LEVEL # finest level of pyramid
assert len(blobs_in) == k_max - k_min + 1
for lvl in range(k_min, k_max + 1):
bl_in = blobs_in[k_max - lvl] # blobs_in is in reversed order
sc = spatial_scales[k_max - lvl] # in reversed order
slvl = str(lvl)
if lvl == k_min:
# Create conv ops with randomly initialized weights and
# zeroed biases for the first FPN level; these will be shared by
# all other FPN levels
# RPN hidden representation
conv_rpn_fpn = model.Conv(
bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
rpn_cls_logits_fpn = model.Conv(
conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
# Proposal bbox regression deltas
rpn_bbox_pred_fpn = model.Conv(
conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0)
)
else:
# Share weights and biases
sk_min = str(k_min)
# RPN hidden representation
conv_rpn_fpn = model.ConvShared(
bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight='conv_rpn_fpn' + sk_min + '_w',
bias='conv_rpn_fpn' + sk_min + '_b'
)
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
rpn_cls_logits_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_cls_logits_fpn' + sk_min + '_w',
bias='rpn_cls_logits_fpn' + sk_min + '_b'
)
# Proposal bbox regression deltas
rpn_bbox_pred_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_bbox_pred_fpn' + sk_min + '_w',
bias='rpn_bbox_pred_fpn' + sk_min + '_b'
)
if not model.train or cfg.MODEL.FASTER_RCNN:
# Proposals are needed during:
# 1) inference (== not model.train) for RPN only and Faster R-CNN
# OR
# 2) training for Faster R-CNN
# Otherwise (== training for RPN only), proposals are not needed
lvl_anchors = generate_anchors(
stride=2.**lvl,
sizes=(cfg.FPN.RPN_ANCHOR_START_SIZE * 2.**(lvl - k_min), ),
aspect_ratios=cfg.FPN.RPN_ASPECT_RATIOS
)
rpn_cls_probs_fpn = model.net.Sigmoid(
rpn_cls_logits_fpn, 'rpn_cls_probs_fpn' + slvl
)
model.GenerateProposals(
[rpn_cls_probs_fpn, rpn_bbox_pred_fpn, 'im_info'],
['rpn_rois_fpn' + slvl, 'rpn_roi_probs_fpn' + slvl],
anchors=lvl_anchors,
spatial_scale=sc
)
def add_fpn_rpn_outputs(model, blobs_in, dim_in, spatial_scales):
"""Add RPN on FPN specific outputs."""
num_anchors = len(cfg.FPN.RPN_ASPECT_RATIOS)
dim_out = dim_in
# 6 2, 一共5层
k_max = cfg.FPN.RPN_MAX_LEVEL # coarsest level of pyramid
k_min = cfg.FPN.RPN_MIN_LEVEL # finest level of pyramid
assert len(blobs_in) == k_max - k_min + 1
for lvl in range(k_min, k_max + 1):
bl_in = blobs_in[k_max - lvl] # blobs_in is in reversed order
sc = spatial_scales[k_max - lvl] # in reversed order
slvl = str(lvl)
# fpn的第一层构建，后面的层共享
if lvl == k_min:
# Create conv ops with randomly initialized weights and
# zeroed biases for the first FPN level; these will be shared by
# all other FPN levels
# RPN hidden representation
# 首先接一个卷积层
conv_rpn_fpn = model.Conv(bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
# 每个anchor包含目标的概率
rpn_cls_logits_fpn = model.Conv(conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
# Proposal bbox regression deltas
# 每个anchor的回归量
rpn_bbox_pred_fpn = model.Conv(conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight_init=gauss_fill(0.01),
bias_init=const_fill(0.0))
else:
# Share weights and biases
sk_min = str(k_min)
# RPN hidden representation
conv_rpn_fpn = model.ConvShared(
bl_in,
'conv_rpn_fpn' + slvl,
dim_in,
dim_out,
kernel=3,
pad=1,
stride=1,
weight='conv_rpn_fpn' + sk_min + '_w',
bias='conv_rpn_fpn' + sk_min + '_b')
model.Relu(conv_rpn_fpn, conv_rpn_fpn)
# Proposal classification scores
rpn_cls_logits_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_cls_logits_fpn' + slvl,
dim_in,
num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_cls_logits_fpn' + sk_min + '_w',
bias='rpn_cls_logits_fpn' + sk_min + '_b')
# Proposal bbox regression deltas
rpn_bbox_pred_fpn = model.ConvShared(
conv_rpn_fpn,
'rpn_bbox_pred_fpn' + slvl,
dim_in,
4 * num_anchors,
kernel=1,
pad=0,
stride=1,
weight='rpn_bbox_pred_fpn' + sk_min + '_w',
bias='rpn_bbox_pred_fpn' + sk_min + '_b')
# 为后面的目标分类和回归提供最好的训练样本
if not model.train or cfg.MODEL.FASTER_RCNN:
# Proposals are needed during:
# 1) inference (== not model.train) for RPN only and Faster R-CNN
# OR
# 2) training for Faster R-CNN
# Otherwise (== training for RPN only), proposals are not needed
# 在进行训练的时候，需要选择区域候选
# 获得本层使用的anchor
lvl_anchors = generate_anchors(
stride=2.**lvl,
sizes=(cfg.FPN.RPN_ANCHOR_START_SIZE * 2.**(lvl - k_min), ),
aspect_ratios=cfg.FPN.RPN_ASPECT_RATIOS)
# slvl是fpn中的层数[2, ..., 6]
# 添加Sigmoid
rpn_cls_probs_fpn = model.net.Sigmoid(rpn_cls_logits_fpn,
'rpn_cls_probs_fpn' + slvl)
# im_info (网络输入的高度， 网络输入的宽度，原始图片到网络输入的放缩比例)
# 获得(R, 5), 每个ROI是相对于网络输入尺寸的矩形
model.GenerateProposals(
[rpn_cls_probs_fpn, rpn_bbox_pred_fpn, 'im_info'], # 输入
['rpn_rois_fpn' + slvl, 'rpn_roi_probs_fpn' + slvl], # 输出
anchors=lvl_anchors,
spatial_scale=sc)
def get_field_of_anchors(
stride, anchor_sizes, anchor_aspect_ratios, octave=None, aspect=None
):
global _threadlocal_foa
if not hasattr(_threadlocal_foa, 'cache'):
_threadlocal_foa.cache = {}
cache_key = str(stride) + str(anchor_sizes) + str(anchor_aspect_ratios)
if cache_key in _threadlocal_foa.cache:
return _threadlocal_foa.cache[cache_key]
# Anchors at a single feature cell
# 一个特征单元格的anchor, cell->格子
cell_anchors = generate_anchors(
stride=stride, sizes=anchor_sizes, aspect_ratios=anchor_aspect_ratios
)
# 3
num_cell_anchors = cell_anchors.shape[0]
# Generate canonical proposals from shifted anchors
# Enumerate all shifted positions on the (H, W) grid
# TRAIN.MAX_SIZE对这里会产生影响
# 在构建fpn时从后往前进行，所以这里先计算最后一层的输出
fpn_max_size = cfg.FPN.COARSEST_STRIDE * np.ceil(
cfg.TRAIN.MAX_SIZE / float(cfg.FPN.COARSEST_STRIDE)
)
# 该层的特征单元的个数
field_size = int(np.ceil(fpn_max_size / float(stride)))
# 有了所有中心点的坐标，在加上anchor的偏移，就得到了RPN
# shifts为对应于网络输入的像素坐标
shifts = np.arange(0, field_size) * stride
# 对应于anchor中心点在网络输入中的坐标集合
shift_x, shift_y = np.meshgrid(shifts, shifts)
shift_x = shift_x.ravel()
shift_y = shift_y.ravel()
# 形状为(., 4)
shifts = np.vstack((shift_x, shift_y, shift_x, shift_y)).transpose()
# Broacast anchors over shifts to enumerate all anchors at all positions
# in the (H, W) grid:
# - add A cell anchors of shape (1, A, 4) to
# - K shifts of shape (K, 1, 4) to get
# - all shifted anchors of shape (K, A, 4)
# - reshape to (K*A, 4) shifted anchors
# anchor的个数
A = num_cell_anchors
# 有多少个位置
K = shifts.shape[0]
field_of_anchors = (
cell_anchors.reshape((1, A, 4)) +
shifts.reshape((1, K, 4)).transpose((1, 0, 2))
)
field_of_anchors = field_of_anchors.reshape((K * A, 4))
# 特征图中所有位置的anchor，以及构造信息
foa = FieldOfAnchors(
field_of_anchors=field_of_anchors.astype(np.float32),
num_cell_anchors=num_cell_anchors,
stride=stride,
field_size=field_size,
octave=octave,
aspect=aspect
)
_threadlocal_foa.cache[cache_key] = foa
return foa
