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
曾经的那一瞬间
Models
提交
e453835a
M
Models
项目概览
曾经的那一瞬间
/
Models
大约 1 年 前同步成功
通知
1
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
DevOps
流水线
流水线任务
计划
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
Models
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
DevOps
DevOps
流水线
流水线任务
计划
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
流水线任务
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
e453835a
编写于
9月 14, 2021
作者:
A
A. Unique TensorFlower
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #10225 from srihari-humbarwadi:coco_tfrecords
PiperOrigin-RevId: 396682327
上级
fec0338f
983ffd16
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
289 addition
and
47 deletion
+289
-47
official/vision/beta/data/create_coco_tf_record.py
official/vision/beta/data/create_coco_tf_record.py
+175
-8
official/vision/beta/data/tfrecord_lib.py
official/vision/beta/data/tfrecord_lib.py
+7
-2
official/vision/beta/projects/panoptic_maskrcnn/configs/panoptic_maskrcnn.py
...a/projects/panoptic_maskrcnn/configs/panoptic_maskrcnn.py
+18
-0
official/vision/beta/projects/panoptic_maskrcnn/dataloaders/panoptic_maskrcnn_input.py
.../panoptic_maskrcnn/dataloaders/panoptic_maskrcnn_input.py
+84
-35
official/vision/beta/projects/panoptic_maskrcnn/tasks/panoptic_maskrcnn.py
...eta/projects/panoptic_maskrcnn/tasks/panoptic_maskrcnn.py
+5
-2
未找到文件。
official/vision/beta/data/create_coco_tf_record.py
浏览文件 @
e453835a
...
...
@@ -58,6 +58,14 @@ flags.DEFINE_string(
'annotations - boxes and instance masks.'
)
flags
.
DEFINE_string
(
'caption_annotations_file'
,
''
,
'File containing image '
'captions.'
)
flags
.
DEFINE_string
(
'panoptic_annotations_file'
,
''
,
'File containing panoptic '
'annotations.'
)
flags
.
DEFINE_string
(
'panoptic_masks_dir'
,
''
,
'Directory containing panoptic masks annotations.'
)
flags
.
DEFINE_boolean
(
'include_panoptic_masks'
,
False
,
'Whether to include category and '
'instance masks in the result. These are required to run the PQ evaluator '
'default: False.'
)
flags
.
DEFINE_string
(
'output_file_prefix'
,
'/tmp/train'
,
'Path to output file'
)
flags
.
DEFINE_integer
(
'num_shards'
,
32
,
'Number of shards for output file.'
)
...
...
@@ -66,6 +74,11 @@ FLAGS = flags.FLAGS
logger
=
tf
.
get_logger
()
logger
.
setLevel
(
logging
.
INFO
)
_VOID_LABEL
=
0
_VOID_INSTANCE_ID
=
0
_THING_CLASS_ID
=
1
_STUFF_CLASSES_OFFSET
=
90
def
coco_segmentation_to_mask_png
(
segmentation
,
height
,
width
,
is_crowd
):
"""Encode a COCO mask segmentation as PNG string."""
...
...
@@ -74,12 +87,79 @@ def coco_segmentation_to_mask_png(segmentation, height, width, is_crowd):
if
not
is_crowd
:
binary_mask
=
np
.
amax
(
binary_mask
,
axis
=
2
)
return
tfrecord_lib
.
encode_binary_mask_as_png
(
binary_mask
)
return
tfrecord_lib
.
encode_mask_as_png
(
binary_mask
)
def
generate_coco_panoptics_masks
(
segments_info
,
mask_path
,
include_panoptic_masks
,
is_category_thing
):
"""Creates masks for panoptic segmentation task.
Args:
segments_info: a list of dicts, where each dict has keys: [u'id',
u'category_id', u'area', u'bbox', u'iscrowd'], detailing information for
each segment in the panoptic mask.
mask_path: path to the panoptic mask.
include_panoptic_masks: bool, when set to True, category and instance
masks are included in the outputs. Set this to True, when using
the Panoptic Quality evaluator.
is_category_thing: a dict with category ids as keys and, 0/1 as values to
represent "stuff" and "things" classes respectively.
Returns:
A dict with with keys: [u'semantic_segmentation_mask', u'category_mask',
u'instance_mask']. The dict contains 'category_mask' and 'instance_mask'
only if `include_panoptic_eval_masks` is set to True.
"""
rgb_mask
=
tfrecord_lib
.
read_image
(
mask_path
)
r
,
g
,
b
=
np
.
split
(
rgb_mask
,
3
,
axis
=-
1
)
# decode rgb encoded panoptic mask to get segments ids
# refer https://cocodataset.org/#format-data
segments_encoded_mask
=
(
r
+
g
*
256
+
b
*
(
256
**
2
)).
squeeze
()
semantic_segmentation_mask
=
np
.
ones_like
(
segments_encoded_mask
,
dtype
=
np
.
uint8
)
*
_VOID_LABEL
if
include_panoptic_masks
:
category_mask
=
np
.
ones_like
(
segments_encoded_mask
,
dtype
=
np
.
uint8
)
*
_VOID_LABEL
instance_mask
=
np
.
ones_like
(
segments_encoded_mask
,
dtype
=
np
.
uint8
)
*
_VOID_INSTANCE_ID
for
idx
,
segment
in
enumerate
(
segments_info
):
segment_id
=
segment
[
'id'
]
category_id
=
segment
[
'category_id'
]
if
is_category_thing
[
category_id
]:
encoded_category_id
=
_THING_CLASS_ID
instance_id
=
idx
+
1
else
:
encoded_category_id
=
category_id
-
_STUFF_CLASSES_OFFSET
instance_id
=
_VOID_INSTANCE_ID
segment_mask
=
(
segments_encoded_mask
==
segment_id
)
semantic_segmentation_mask
[
segment_mask
]
=
encoded_category_id
if
include_panoptic_masks
:
category_mask
[
segment_mask
]
=
category_id
instance_mask
[
segment_mask
]
=
instance_id
outputs
=
{
'semantic_segmentation_mask'
:
tfrecord_lib
.
encode_mask_as_png
(
semantic_segmentation_mask
)
}
if
include_panoptic_masks
:
outputs
.
update
({
'category_mask'
:
tfrecord_lib
.
encode_mask_as_png
(
category_mask
),
'instance_mask'
:
tfrecord_lib
.
encode_mask_as_png
(
instance_mask
)
})
return
outputs
def
coco_annotations_to_lists
(
bbox_annotations
,
id_to_name_map
,
image_height
,
image_width
,
include_masks
):
"""Convert COCO annotations to feature lists."""
"""Convert
s
COCO annotations to feature lists."""
data
=
dict
((
k
,
list
())
for
k
in
[
'xmin'
,
'xmax'
,
'ymin'
,
'ymax'
,
'is_crowd'
,
...
...
@@ -160,9 +240,13 @@ def encode_caption_annotations(caption_annotations):
def
create_tf_example
(
image
,
image_dirs
,
panoptic_masks_dir
=
None
,
bbox_annotations
=
None
,
id_to_name_map
=
None
,
caption_annotations
=
None
,
panoptic_annotation
=
None
,
is_category_thing
=
None
,
include_panoptic_masks
=
False
,
include_masks
=
False
):
"""Converts image and annotations to a tf.Example proto.
...
...
@@ -170,6 +254,7 @@ def create_tf_example(image,
image: dict with keys: [u'license', u'file_name', u'coco_url', u'height',
u'width', u'date_captured', u'flickr_url', u'id']
image_dirs: list of directories containing the image files.
panoptic_masks_dir: `str` of the panoptic masks directory.
bbox_annotations:
list of dicts with keys: [u'segmentation', u'area', u'iscrowd',
u'image_id', u'bbox', u'category_id', u'id'] Notice that bounding box
...
...
@@ -182,6 +267,11 @@ def create_tf_example(image,
id_to_name_map: a dict mapping category IDs to string names.
caption_annotations:
list of dict with keys: [u'id', u'image_id', u'str'].
panoptic_annotation: dict with keys: [u'image_id', u'file_name',
u'segments_info']. Where the value for segments_info is a list of dicts,
with each dict containing information for a single segment in the mask.
is_category_thing: `bool`, whether it is a category thing.
include_panoptic_masks: `bool`, whether to include panoptic masks.
include_masks: Whether to include instance segmentations masks
(PNG encoded) in the result. default: False.
...
...
@@ -234,6 +324,26 @@ def create_tf_example(image,
feature_dict
.
update
(
{
'image/caption'
:
tfrecord_lib
.
convert_to_feature
(
encoded_captions
)})
if
panoptic_annotation
:
segments_info
=
panoptic_annotation
[
'segments_info'
]
panoptic_mask_filename
=
os
.
path
.
join
(
panoptic_masks_dir
,
panoptic_annotation
[
'file_name'
])
encoded_panoptic_masks
=
generate_coco_panoptics_masks
(
segments_info
,
panoptic_mask_filename
,
include_panoptic_masks
,
is_category_thing
)
feature_dict
.
update
(
{
'image/segmentation/class/encoded'
:
tfrecord_lib
.
convert_to_feature
(
encoded_panoptic_masks
[
'semantic_segmentation_mask'
])})
if
include_panoptic_masks
:
feature_dict
.
update
({
'image/panoptic/category_mask'
:
tfrecord_lib
.
convert_to_feature
(
encoded_panoptic_masks
[
'category_mask'
]),
'image/panoptic/instance_mask'
:
tfrecord_lib
.
convert_to_feature
(
encoded_panoptic_masks
[
'instance_mask'
])
})
example
=
tf
.
train
.
Example
(
features
=
tf
.
train
.
Features
(
feature
=
feature_dict
))
return
example
,
num_annotations_skipped
...
...
@@ -287,6 +397,33 @@ def _load_caption_annotations(caption_annotations_file):
return
img_to_caption_annotation
def
_load_panoptic_annotations
(
panoptic_annotations_file
):
"""Loads panoptic annotation from file."""
with
tf
.
io
.
gfile
.
GFile
(
panoptic_annotations_file
,
'r'
)
as
fid
:
panoptic_annotations
=
json
.
load
(
fid
)
img_to_panoptic_annotation
=
dict
()
logging
.
info
(
'Building panoptic index.'
)
for
annotation
in
panoptic_annotations
[
'annotations'
]:
image_id
=
annotation
[
'image_id'
]
img_to_panoptic_annotation
[
image_id
]
=
annotation
is_category_thing
=
dict
()
for
category_info
in
panoptic_annotations
[
'categories'
]:
is_category_thing
[
category_info
[
'id'
]]
=
category_info
[
'isthing'
]
==
1
missing_annotation_count
=
0
images
=
panoptic_annotations
[
'images'
]
for
image
in
images
:
image_id
=
image
[
'id'
]
if
image_id
not
in
img_to_panoptic_annotation
:
missing_annotation_count
+=
1
logging
.
info
(
'%d images are missing panoptic annotations.'
,
missing_annotation_count
)
return
img_to_panoptic_annotation
,
is_category_thing
def
_load_images_info
(
images_info_file
):
with
tf
.
io
.
gfile
.
GFile
(
images_info_file
,
'r'
)
as
fid
:
info_dict
=
json
.
load
(
fid
)
...
...
@@ -294,11 +431,15 @@ def _load_images_info(images_info_file):
def
generate_annotations
(
images
,
image_dirs
,
panoptic_masks_dir
=
None
,
img_to_obj_annotation
=
None
,
img_to_caption_annotation
=
None
,
id_to_name_map
=
None
,
img_to_caption_annotation
=
None
,
img_to_panoptic_annotation
=
None
,
is_category_thing
=
None
,
id_to_name_map
=
None
,
include_panoptic_masks
=
False
,
include_masks
=
False
):
"""Generator for COCO annotations."""
for
image
in
images
:
object_annotation
=
(
img_to_obj_annotation
.
get
(
image
[
'id'
],
None
)
if
img_to_obj_annotation
else
None
)
...
...
@@ -306,8 +447,11 @@ def generate_annotations(images, image_dirs,
caption_annotaion
=
(
img_to_caption_annotation
.
get
(
image
[
'id'
],
None
)
if
img_to_caption_annotation
else
None
)
yield
(
image
,
image_dirs
,
object_annotation
,
id_to_name_map
,
caption_annotaion
,
include_masks
)
panoptic_annotation
=
(
img_to_panoptic_annotation
.
get
(
image
[
'id'
],
None
)
if
img_to_panoptic_annotation
else
None
)
yield
(
image
,
image_dirs
,
panoptic_masks_dir
,
object_annotation
,
id_to_name_map
,
caption_annotaion
,
panoptic_annotation
,
is_category_thing
,
include_panoptic_masks
,
include_masks
)
def
_create_tf_record_from_coco_annotations
(
images_info_file
,
...
...
@@ -316,6 +460,9 @@ def _create_tf_record_from_coco_annotations(images_info_file,
num_shards
,
object_annotations_file
=
None
,
caption_annotations_file
=
None
,
panoptic_masks_dir
=
None
,
panoptic_annotations_file
=
None
,
include_panoptic_masks
=
False
,
include_masks
=
False
):
"""Loads COCO annotation json files and converts to tf.Record format.
...
...
@@ -331,6 +478,10 @@ def _create_tf_record_from_coco_annotations(images_info_file,
num_shards: Number of output files to create.
object_annotations_file: JSON file containing bounding box annotations.
caption_annotations_file: JSON file containing caption annotations.
panoptic_masks_dir: Directory containing panoptic masks.
panoptic_annotations_file: JSON file containing panoptic annotations.
include_panoptic_masks: Whether to include 'category_mask'
and 'instance_mask', which is required by the panoptic quality evaluator.
include_masks: Whether to include instance segmentations masks
(PNG encoded) in the result. default: False.
"""
...
...
@@ -342,16 +493,29 @@ def _create_tf_record_from_coco_annotations(images_info_file,
img_to_obj_annotation
=
None
img_to_caption_annotation
=
None
id_to_name_map
=
None
img_to_panoptic_annotation
=
None
is_category_thing
=
None
if
object_annotations_file
:
img_to_obj_annotation
,
id_to_name_map
=
(
_load_object_annotations
(
object_annotations_file
))
if
caption_annotations_file
:
img_to_caption_annotation
=
(
_load_caption_annotations
(
caption_annotations_file
))
if
panoptic_annotations_file
:
img_to_panoptic_annotation
,
is_category_thing
=
(
_load_panoptic_annotations
(
panoptic_annotations_file
))
coco_annotations_iter
=
generate_annotations
(
images
,
image_dirs
,
img_to_obj_annotation
,
img_to_caption_annotation
,
id_to_name_map
=
id_to_name_map
,
include_masks
=
include_masks
)
images
=
images
,
image_dirs
=
image_dirs
,
panoptic_masks_dir
=
panoptic_masks_dir
,
img_to_obj_annotation
=
img_to_obj_annotation
,
img_to_caption_annotation
=
img_to_caption_annotation
,
img_to_panoptic_annotation
=
img_to_panoptic_annotation
,
is_category_thing
=
is_category_thing
,
id_to_name_map
=
id_to_name_map
,
include_panoptic_masks
=
include_panoptic_masks
,
include_masks
=
include_masks
)
num_skipped
=
tfrecord_lib
.
write_tf_record_dataset
(
output_path
,
coco_annotations_iter
,
create_tf_example
,
num_shards
)
...
...
@@ -380,6 +544,9 @@ def main(_):
FLAGS
.
num_shards
,
FLAGS
.
object_annotations_file
,
FLAGS
.
caption_annotations_file
,
FLAGS
.
panoptic_masks_dir
,
FLAGS
.
panoptic_annotations_file
,
FLAGS
.
include_panoptic_masks
,
FLAGS
.
include_masks
)
...
...
official/vision/beta/data/tfrecord_lib.py
浏览文件 @
e453835a
...
...
@@ -100,8 +100,13 @@ def image_info_to_feature_dict(height, width, filename, image_id,
}
def
encode_binary_mask_as_png
(
binary_mask
):
pil_image
=
Image
.
fromarray
(
binary_mask
)
def
read_image
(
image_path
):
pil_image
=
Image
.
open
(
image_path
)
return
np
.
asarray
(
pil_image
)
def
encode_mask_as_png
(
mask
):
pil_image
=
Image
.
fromarray
(
mask
)
output_io
=
io
.
BytesIO
()
pil_image
.
save
(
output_io
,
format
=
'PNG'
)
return
output_io
.
getvalue
()
...
...
official/vision/beta/projects/panoptic_maskrcnn/configs/panoptic_maskrcnn.py
浏览文件 @
e453835a
...
...
@@ -21,6 +21,7 @@ from typing import List, Optional
from
official.core
import
config_definitions
as
cfg
from
official.core
import
exp_factory
from
official.modeling
import
optimization
from
official.vision.beta.configs
import
common
from
official.vision.beta.configs
import
maskrcnn
from
official.vision.beta.configs
import
semantic_segmentation
...
...
@@ -46,11 +47,28 @@ class Parser(maskrcnn.Parser):
segmentation_groundtruth_padded_size
:
List
[
int
]
=
dataclasses
.
field
(
default_factory
=
list
)
segmentation_ignore_label
:
int
=
255
panoptic_ignore_label
:
int
=
0
# Setting this to true will enable parsing category_mask and instance_mask.
include_panoptic_masks
:
bool
=
True
@
dataclasses
.
dataclass
class
TfExampleDecoder
(
common
.
TfExampleDecoder
):
"""A simple TF Example decoder config."""
# Setting this to true will enable decoding category_mask and instance_mask.
include_panoptic_masks
:
bool
=
True
@
dataclasses
.
dataclass
class
DataDecoder
(
common
.
DataDecoder
):
"""Data decoder config."""
simple_decoder
:
TfExampleDecoder
=
TfExampleDecoder
()
@
dataclasses
.
dataclass
class
DataConfig
(
maskrcnn
.
DataConfig
):
"""Input config for training."""
decoder
:
DataDecoder
=
DataDecoder
()
parser
:
Parser
=
Parser
()
...
...
official/vision/beta/projects/panoptic_maskrcnn/dataloaders/panoptic_maskrcnn_input.py
浏览文件 @
e453835a
...
...
@@ -24,25 +24,51 @@ from official.vision.beta.ops import preprocess_ops
class
TfExampleDecoder
(
tf_example_decoder
.
TfExampleDecoder
):
"""Tensorflow Example proto decoder."""
def
__init__
(
self
,
regenerate_source_id
,
mask_binarize_threshold
):
def
__init__
(
self
,
regenerate_source_id
,
mask_binarize_threshold
,
include_panoptic_masks
):
super
(
TfExampleDecoder
,
self
).
__init__
(
include_mask
=
True
,
regenerate_source_id
=
regenerate_source_id
,
mask_binarize_threshold
=
None
)
self
.
_segmentation_keys_to_features
=
{
self
.
_include_panoptic_masks
=
include_panoptic_masks
keys_to_features
=
{
'image/segmentation/class/encoded'
:
tf
.
io
.
FixedLenFeature
((),
tf
.
string
,
default_value
=
''
)
}
tf
.
io
.
FixedLenFeature
((),
tf
.
string
,
default_value
=
''
)}
if
include_panoptic_masks
:
keys_to_features
.
update
({
'image/panoptic/category_mask'
:
tf
.
io
.
FixedLenFeature
((),
tf
.
string
,
default_value
=
''
),
'image/panoptic/instance_mask'
:
tf
.
io
.
FixedLenFeature
((),
tf
.
string
,
default_value
=
''
)})
self
.
_segmentation_keys_to_features
=
keys_to_features
def
decode
(
self
,
serialized_example
):
decoded_tensors
=
super
(
TfExampleDecoder
,
self
).
decode
(
serialized_example
)
segmentation_
parsed_tensors
=
tf
.
io
.
parse_single_example
(
parsed_tensors
=
tf
.
io
.
parse_single_example
(
serialized_example
,
self
.
_segmentation_keys_to_features
)
segmentation_mask
=
tf
.
io
.
decode_image
(
segmentation_
parsed_tensors
[
'image/segmentation/class/encoded'
],
parsed_tensors
[
'image/segmentation/class/encoded'
],
channels
=
1
)
segmentation_mask
.
set_shape
([
None
,
None
,
1
])
decoded_tensors
.
update
({
'groundtruth_segmentation_mask'
:
segmentation_mask
})
if
self
.
_include_panoptic_masks
:
category_mask
=
tf
.
io
.
decode_image
(
parsed_tensors
[
'image/panoptic/category_mask'
],
channels
=
1
)
instance_mask
=
tf
.
io
.
decode_image
(
parsed_tensors
[
'image/panoptic/instance_mask'
],
channels
=
1
)
category_mask
.
set_shape
([
None
,
None
,
1
])
instance_mask
.
set_shape
([
None
,
None
,
1
])
decoded_tensors
.
update
({
'groundtruth_panoptic_category_mask'
:
category_mask
,
'groundtruth_panoptic_instance_mask'
:
instance_mask
})
return
decoded_tensors
...
...
@@ -69,6 +95,8 @@ class Parser(maskrcnn_input.Parser):
segmentation_resize_eval_groundtruth
=
True
,
segmentation_groundtruth_padded_size
=
None
,
segmentation_ignore_label
=
255
,
panoptic_ignore_label
=
0
,
include_panoptic_masks
=
True
,
dtype
=
'float32'
):
"""Initializes parameters for parsing annotations in the dataset.
...
...
@@ -106,8 +134,12 @@ class Parser(maskrcnn_input.Parser):
segmentation_groundtruth_padded_size: `Tensor` or `list` for [height,
width]. When resize_eval_groundtruth is set to False, the groundtruth
masks are padded to this size.
segmentation_ignore_label: `int` the pixel with ignore label will not used
for training and evaluation.
segmentation_ignore_label: `int` the pixels with ignore label will not be
used for training and evaluation.
panoptic_ignore_label: `int` the pixels with ignore label will not be used
by the PQ evaluator.
include_panoptic_masks: `bool`, if True, category_mask and instance_mask
will be parsed. Set this to true if PQ evaluator is enabled.
dtype: `str`, data type. One of {`bfloat16`, `float32`, `float16`}.
"""
super
(
Parser
,
self
).
__init__
(
...
...
@@ -139,6 +171,8 @@ class Parser(maskrcnn_input.Parser):
'specified when segmentation_resize_eval_groundtruth is False.'
)
self
.
_segmentation_groundtruth_padded_size
=
segmentation_groundtruth_padded_size
self
.
_segmentation_ignore_label
=
segmentation_ignore_label
self
.
_panoptic_ignore_label
=
panoptic_ignore_label
self
.
_include_panoptic_masks
=
include_panoptic_masks
def
_parse_train_data
(
self
,
data
):
"""Parses data for training.
...
...
@@ -250,39 +284,54 @@ class Parser(maskrcnn_input.Parser):
shape [height_l, width_l, 4] representing anchor boxes at each
level.
"""
segmentation_mask
=
tf
.
cast
(
data
[
'groundtruth_segmentation_mask'
],
tf
.
float32
)
segmentation_mask
=
tf
.
reshape
(
segmentation_mask
,
shape
=
[
1
,
data
[
'height'
],
data
[
'width'
],
1
])
segmentation_mask
+=
1
def
_process_mask
(
mask
,
ignore_label
,
image_info
):
mask
=
tf
.
cast
(
mask
,
dtype
=
tf
.
float32
)
mask
=
tf
.
reshape
(
mask
,
shape
=
[
1
,
data
[
'height'
],
data
[
'width'
],
1
])
mask
+=
1
if
self
.
_segmentation_resize_eval_groundtruth
:
# Resizes eval masks to match input image sizes. In that case, mean IoU
# is computed on output_size not the original size of the images.
image_scale
=
image_info
[
2
,
:]
offset
=
image_info
[
3
,
:]
mask
=
preprocess_ops
.
resize_and_crop_masks
(
mask
,
image_scale
,
self
.
_output_size
,
offset
)
else
:
mask
=
tf
.
image
.
pad_to_bounding_box
(
mask
,
0
,
0
,
self
.
_segmentation_groundtruth_padded_size
[
0
],
self
.
_segmentation_groundtruth_padded_size
[
1
])
mask
-=
1
# Assign ignore label to the padded region.
mask
=
tf
.
where
(
tf
.
equal
(
mask
,
-
1
),
ignore_label
*
tf
.
ones_like
(
mask
),
mask
)
mask
=
tf
.
squeeze
(
mask
,
axis
=
0
)
return
mask
image
,
labels
=
super
(
Parser
,
self
).
_parse_eval_data
(
data
)
image_info
=
labels
[
'image_info'
]
if
self
.
_segmentation_resize_eval_groundtruth
:
# Resizes eval masks to match input image sizes. In that case, mean IoU
# is computed on output_size not the original size of the images.
image_info
=
labels
[
'image_info'
]
image_scale
=
image_info
[
2
,
:]
offset
=
image_info
[
3
,
:]
segmentation_mask
=
preprocess_ops
.
resize_and_crop_masks
(
segmentation_mask
,
image_scale
,
self
.
_output_size
,
offset
)
else
:
segmentation_mask
=
tf
.
image
.
pad_to_bounding_box
(
segmentation_mask
,
0
,
0
,
self
.
_segmentation_groundtruth_padded_size
[
0
],
self
.
_segmentation_groundtruth_padded_size
[
1
])
segmentation_mask
-=
1
# Assign ignore label to the padded region.
segmentation_mask
=
tf
.
where
(
tf
.
equal
(
segmentation_mask
,
-
1
),
self
.
_segmentation_ignore_label
*
tf
.
ones_like
(
segmentation_mask
),
segmentation_mask
)
segmentation_mask
=
tf
.
squeeze
(
segmentation_mask
,
axis
=
0
)
segmentation_mask
=
_process_mask
(
data
[
'groundtruth_segmentation_mask'
],
self
.
_segmentation_ignore_label
,
image_info
)
segmentation_valid_mask
=
tf
.
not_equal
(
segmentation_mask
,
self
.
_segmentation_ignore_label
)
labels
[
'groundtruths'
].
update
({
'gt_segmentation_mask'
:
segmentation_mask
,
'gt_segmentation_valid_mask'
:
segmentation_valid_mask
})
if
self
.
_include_panoptic_masks
:
panoptic_category_mask
=
_process_mask
(
data
[
'groundtruth_panoptic_category_mask'
],
self
.
_panoptic_ignore_label
,
image_info
)
panoptic_instance_mask
=
_process_mask
(
data
[
'groundtruth_panoptic_instance_mask'
],
self
.
_panoptic_ignore_label
,
image_info
)
labels
[
'groundtruths'
].
update
({
'gt_panoptic_category_mask'
:
panoptic_category_mask
,
'gt_panoptic_instance_mask'
:
panoptic_instance_mask
})
return
image
,
labels
official/vision/beta/projects/panoptic_maskrcnn/tasks/panoptic_maskrcnn.py
浏览文件 @
e453835a
...
...
@@ -121,7 +121,8 @@ class PanopticMaskRCNNTask(maskrcnn.MaskRCNNTask):
if
params
.
decoder
.
type
==
'simple_decoder'
:
decoder
=
panoptic_maskrcnn_input
.
TfExampleDecoder
(
regenerate_source_id
=
decoder_cfg
.
regenerate_source_id
,
mask_binarize_threshold
=
decoder_cfg
.
mask_binarize_threshold
)
mask_binarize_threshold
=
decoder_cfg
.
mask_binarize_threshold
,
include_panoptic_masks
=
decoder_cfg
.
include_panoptic_masks
)
else
:
raise
ValueError
(
'Unknown decoder type: {}!'
.
format
(
params
.
decoder
.
type
))
...
...
@@ -147,7 +148,9 @@ class PanopticMaskRCNNTask(maskrcnn.MaskRCNNTask):
.
segmentation_resize_eval_groundtruth
,
segmentation_groundtruth_padded_size
=
params
.
parser
.
segmentation_groundtruth_padded_size
,
segmentation_ignore_label
=
params
.
parser
.
segmentation_ignore_label
)
segmentation_ignore_label
=
params
.
parser
.
segmentation_ignore_label
,
panoptic_ignore_label
=
params
.
parser
.
panoptic_ignore_label
,
include_panoptic_masks
=
params
.
parser
.
include_panoptic_masks
)
reader
=
input_reader_factory
.
input_reader_generator
(
params
,
...
...
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