Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
27号BigBang
Mask_RCNN
提交
166a215f
M
Mask_RCNN
项目概览
27号BigBang
/
Mask_RCNN
与 Fork 源项目一致
从无法访问的项目Fork
通知
3
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
Mask_RCNN
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
前往新版Gitcode,体验更适合开发者的 AI 搜索 >>
提交
166a215f
编写于
1月 27, 2018
作者:
S
Shenoy
提交者:
Cory Pruce
1月 28, 2018
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
adding changes by LandyGuo
上级
032d9eb4
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
40 addition
and
32 deletion
+40
-32
model.py
model.py
+40
-32
未找到文件。
model.py
浏览文件 @
166a215f
...
...
@@ -782,53 +782,61 @@ def refine_detections_graph(rois, probs, deltas, window, config):
# Filter out background boxes
keep
=
tf
.
where
(
class_ids
>
0
)[:,
0
]
# Filter out low confidence boxes
conf_keep
=
tf
.
where
(
class_scores
>=
config
.
DETECTION_MIN_CONFIDENCE
)[:,
0
]
if
config
.
DETECTION_MIN_CONFIDENCE
:
keep
=
tf
.
sets
.
set_intersection
(
keep
,
tf
.
where
(
class_scores
>=
config
.
DETECTION_MIN_CONFIDENCE
))[:,
0
]
keep
=
tf
.
sparse_tensor_to_dense
(
tf
.
sets
.
set_intersection
(
tf
.
expand_dims
(
keep
,
0
),
tf
.
expand_dims
(
conf_keep
,
0
)))[
0
]
# Apply per-class NMS
pre_nms_class_ids
=
tf
.
gather
(
class_ids
,
keep
)
pre_nms_scores
=
tf
.
gather
(
class_scores
,
keep
)
pre_nms_rois
=
tf
.
gather
(
refined_rois
,
keep
)
print
(
'pre_nms_class_ids = {}'
.
format
(
pre_nms_class_ids
.
shape
))
print
(
'pre_nms_scores = {}'
.
format
(
pre_nms_scores
.
shape
))
print
(
'pre_nms_rois = {}'
.
format
(
pre_nms_rois
.
shape
))
uniq_pre_nms_class_ids
=
tf
.
unique
(
pre_nms_class_ids
)[
0
]
# sort unique class ids
_
,
max_index
=
tf
.
nn
.
top_k
(
-
uniq_pre_nms_class_ids
,
tf
.
size
(
uniq_pre_nms_class_ids
))
uniq_pre_nms_class_ids
=
tf
.
gather
(
uniq_pre_nms_class_ids
,
max_index
)
nms_keep
=
[]
def
nms_keep_map
(
class_id
):
#class_id = tf.expand_dims(class_id, -1)
print
(
'pre_nms_class_ids.shape'
,
pre_nms_class_ids
.
shape
)
print
(
'class_id'
,
class_id
.
shape
)
ixs
=
tf
.
where
(
pre_nms_class_ids
==
class_id
)
def
nms_keep_map
(
i
,
ret
):
class_id
=
uniq_pre_nms_class_ids
[
i
]
scale
=
tf
.
fill
(
tf
.
shape
(
pre_nms_class_ids
),
class_id
)
ixs
=
tf
.
cast
(
tf
.
where
(
tf
.
equal
(
scale
,
pre_nms_class_ids
))[:,
0
],
tf
.
int32
)
print
(
'ixs {}'
.
format
(
ixs
.
shape
))
# Apply NMS
class_keep
=
tf
.
image
.
non_max_suppression
(
tf
.
to_float
(
tf
.
gather
(
pre_nms_rois
,
ixs
)),
tf
.
gather
(
pre_nms_scores
,
ixs
),
max_output_size
=
ixs
.
shape
[
1
],
max_output_size
=
tf
.
shape
(
ixs
)[
0
],
iou_threshold
=
config
.
DETECTION_NMS_THRESHOLD
)
# Map indicies
return
tf
.
gather
(
keep
,
tf
.
gather
(
ixs
,
class_keep
))
print
(
'uniq_pre_nms_class_ids: {}'
.
format
(
uniq_pre_nms_class_ids
.
shape
))
nms_keep
=
tf
.
map_fn
(
nms_keep_map
,
uniq_pre_nms_class_ids
)
nms_keep
=
tf
.
concat
(
nms_keep
,
axis
=
0
)
nms_keep
=
tf
.
unique
(
nms_keep
)[
0
]
nms_keep
=
tf
.
to_int64
(
nms_keep
)
keep
=
tf
.
expand_dims
(
keep
,
0
)
nms_keep
=
tf
.
expand_dims
(
nms_keep
,
0
)
keep
=
tf
.
sparse_tensor_to_dense
(
tf
.
sets
.
set_intersection
(
keep
,
nms_keep
))[:,
0
]
"""#tf.to_int32(
#np.intersect1d(keep, nms_keep).astype(np.int32)
"""
cur_keep_indexes
=
tf
.
gather
(
keep
,
tf
.
gather
(
ixs
,
class_keep
))
return
i
+
1
,
tf
.
concat
([
ret
,
cur_keep_indexes
],
axis
=
0
)
nums_iters
=
tf
.
shape
(
uniq_pre_nms_class_ids
)[
0
]
# unique class ids
i
=
tf
.
constant
(
0
)
ret
=
tf
.
ones
([
1
],
dtype
=
tf
.
int32
)
c
=
lambda
i
,
unique_pre_nms
:
tf
.
less
(
i
,
nums_iters
)
b
=
nms_keep_map
r
=
tf
.
while_loop
(
c
,
b
,
[
i
,
-
ret
],
shape_invariants
=
[
i
.
get_shape
(),
tf
.
TensorShape
([
None
])])
nms_keep
=
r
[
1
]
# remove initial_value background
nms_keep
=
tf
.
gather
(
nms_keep
,
tf
.
where
(
nms_keep
>=
0
)[:,
0
])
keep
=
tf
.
sparse_tensor_to_dense
(
tf
.
sets
.
set_intersection
(
tf
.
expand_dims
(
keep
,
0
),
tf
.
expand_dims
(
nms_keep
,
0
)))[
0
]
keep
=
tf
.
cast
(
keep
,
tf
.
int32
)
# Keep top detections
roi_count
=
tf
.
convert_to_tensor
(
config
.
DETECTION_MAX_INSTANCES
)
print
(
'class scores: {}'
.
format
(
class_scores
.
shape
))
class_scores_keep
=
tf
.
gather
(
class_scores
,
keep
)
num_keep
=
tf
.
minimum
(
tf
.
shape
(
class_scores_keep
)[
0
],
roi_count
)
top_ids
=
tf
.
nn
.
top_k
(
class_scores_keep
,
k
=
num_keep
,
sorted
=
True
)[
1
]
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录