2020-12-09 14:45:25 +08:00
|
|
|
# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
|
|
|
|
#
|
|
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
# you may not use this file except in compliance with the License.
|
|
|
|
# You may obtain a copy of the License at
|
|
|
|
#
|
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
#
|
|
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
# See the License for the specific language governing permissions and
|
|
|
|
# limitations under the License.
|
|
|
|
|
|
|
|
from __future__ import absolute_import
|
|
|
|
from __future__ import division
|
|
|
|
from __future__ import print_function
|
|
|
|
|
|
|
|
import paddle
|
|
|
|
from paddle import nn
|
|
|
|
from .det_basic_loss import DiceLoss
|
|
|
|
import numpy as np
|
|
|
|
|
|
|
|
|
|
|
|
class SASTLoss(nn.Layer):
|
|
|
|
"""
|
|
|
|
"""
|
|
|
|
|
2020-12-21 17:13:32 +08:00
|
|
|
def __init__(self, eps=1e-6, **kwargs):
|
2020-12-09 14:45:25 +08:00
|
|
|
super(SASTLoss, self).__init__()
|
|
|
|
self.dice_loss = DiceLoss(eps=eps)
|
|
|
|
|
|
|
|
def forward(self, predicts, labels):
|
|
|
|
"""
|
|
|
|
tcl_pos: N x 128 x 3
|
|
|
|
tcl_mask: N x 128 x 1
|
|
|
|
tcl_label: N x X list or LoDTensor
|
|
|
|
"""
|
2020-12-21 17:13:32 +08:00
|
|
|
|
2020-12-09 14:45:25 +08:00
|
|
|
f_score = predicts['f_score']
|
|
|
|
f_border = predicts['f_border']
|
|
|
|
f_tvo = predicts['f_tvo']
|
|
|
|
f_tco = predicts['f_tco']
|
|
|
|
|
|
|
|
l_score, l_border, l_mask, l_tvo, l_tco = labels[1:]
|
|
|
|
|
|
|
|
#score_loss
|
|
|
|
intersection = paddle.sum(f_score * l_score * l_mask)
|
|
|
|
union = paddle.sum(f_score * l_mask) + paddle.sum(l_score * l_mask)
|
|
|
|
score_loss = 1.0 - 2 * intersection / (union + 1e-5)
|
|
|
|
|
|
|
|
#border loss
|
2020-12-21 17:13:32 +08:00
|
|
|
l_border_split, l_border_norm = paddle.split(
|
|
|
|
l_border, num_or_sections=[4, 1], axis=1)
|
2020-12-09 14:45:25 +08:00
|
|
|
f_border_split = f_border
|
|
|
|
border_ex_shape = l_border_norm.shape * np.array([1, 4, 1, 1])
|
2020-12-21 17:13:32 +08:00
|
|
|
l_border_norm_split = paddle.expand(
|
|
|
|
x=l_border_norm, shape=border_ex_shape)
|
|
|
|
l_border_score = paddle.expand(x=l_score, shape=border_ex_shape)
|
|
|
|
l_border_mask = paddle.expand(x=l_mask, shape=border_ex_shape)
|
2020-12-09 14:45:25 +08:00
|
|
|
|
|
|
|
border_diff = l_border_split - f_border_split
|
2020-12-21 17:13:32 +08:00
|
|
|
abs_border_diff = paddle.abs(border_diff)
|
2020-12-09 14:45:25 +08:00
|
|
|
border_sign = abs_border_diff < 1.0
|
|
|
|
border_sign = paddle.cast(border_sign, dtype='float32')
|
|
|
|
border_sign.stop_gradient = True
|
|
|
|
border_in_loss = 0.5 * abs_border_diff * abs_border_diff * border_sign + \
|
|
|
|
(abs_border_diff - 0.5) * (1.0 - border_sign)
|
|
|
|
border_out_loss = l_border_norm_split * border_in_loss
|
|
|
|
border_loss = paddle.sum(border_out_loss * l_border_score * l_border_mask) / \
|
|
|
|
(paddle.sum(l_border_score * l_border_mask) + 1e-5)
|
|
|
|
|
|
|
|
#tvo_loss
|
2020-12-21 17:13:32 +08:00
|
|
|
l_tvo_split, l_tvo_norm = paddle.split(
|
|
|
|
l_tvo, num_or_sections=[8, 1], axis=1)
|
2020-12-09 14:45:25 +08:00
|
|
|
f_tvo_split = f_tvo
|
|
|
|
tvo_ex_shape = l_tvo_norm.shape * np.array([1, 8, 1, 1])
|
|
|
|
l_tvo_norm_split = paddle.expand(x=l_tvo_norm, shape=tvo_ex_shape)
|
2020-12-21 17:13:32 +08:00
|
|
|
l_tvo_score = paddle.expand(x=l_score, shape=tvo_ex_shape)
|
|
|
|
l_tvo_mask = paddle.expand(x=l_mask, shape=tvo_ex_shape)
|
2020-12-09 14:45:25 +08:00
|
|
|
#
|
|
|
|
tvo_geo_diff = l_tvo_split - f_tvo_split
|
2020-12-21 17:13:32 +08:00
|
|
|
abs_tvo_geo_diff = paddle.abs(tvo_geo_diff)
|
2020-12-09 14:45:25 +08:00
|
|
|
tvo_sign = abs_tvo_geo_diff < 1.0
|
|
|
|
tvo_sign = paddle.cast(tvo_sign, dtype='float32')
|
|
|
|
tvo_sign.stop_gradient = True
|
|
|
|
tvo_in_loss = 0.5 * abs_tvo_geo_diff * abs_tvo_geo_diff * tvo_sign + \
|
|
|
|
(abs_tvo_geo_diff - 0.5) * (1.0 - tvo_sign)
|
|
|
|
tvo_out_loss = l_tvo_norm_split * tvo_in_loss
|
|
|
|
tvo_loss = paddle.sum(tvo_out_loss * l_tvo_score * l_tvo_mask) / \
|
|
|
|
(paddle.sum(l_tvo_score * l_tvo_mask) + 1e-5)
|
|
|
|
|
|
|
|
#tco_loss
|
2020-12-21 17:13:32 +08:00
|
|
|
l_tco_split, l_tco_norm = paddle.split(
|
|
|
|
l_tco, num_or_sections=[2, 1], axis=1)
|
2020-12-09 14:45:25 +08:00
|
|
|
f_tco_split = f_tco
|
|
|
|
tco_ex_shape = l_tco_norm.shape * np.array([1, 2, 1, 1])
|
|
|
|
l_tco_norm_split = paddle.expand(x=l_tco_norm, shape=tco_ex_shape)
|
2020-12-21 17:13:32 +08:00
|
|
|
l_tco_score = paddle.expand(x=l_score, shape=tco_ex_shape)
|
|
|
|
l_tco_mask = paddle.expand(x=l_mask, shape=tco_ex_shape)
|
|
|
|
|
2020-12-09 14:45:25 +08:00
|
|
|
tco_geo_diff = l_tco_split - f_tco_split
|
2020-12-21 17:13:32 +08:00
|
|
|
abs_tco_geo_diff = paddle.abs(tco_geo_diff)
|
2020-12-09 14:45:25 +08:00
|
|
|
tco_sign = abs_tco_geo_diff < 1.0
|
|
|
|
tco_sign = paddle.cast(tco_sign, dtype='float32')
|
|
|
|
tco_sign.stop_gradient = True
|
|
|
|
tco_in_loss = 0.5 * abs_tco_geo_diff * abs_tco_geo_diff * tco_sign + \
|
|
|
|
(abs_tco_geo_diff - 0.5) * (1.0 - tco_sign)
|
|
|
|
tco_out_loss = l_tco_norm_split * tco_in_loss
|
|
|
|
tco_loss = paddle.sum(tco_out_loss * l_tco_score * l_tco_mask) / \
|
|
|
|
(paddle.sum(l_tco_score * l_tco_mask) + 1e-5)
|
|
|
|
|
|
|
|
# total loss
|
|
|
|
tvo_lw, tco_lw = 1.5, 1.5
|
|
|
|
score_lw, border_lw = 1.0, 1.0
|
|
|
|
total_loss = score_loss * score_lw + border_loss * border_lw + \
|
|
|
|
tvo_loss * tvo_lw + tco_loss * tco_lw
|
2020-12-21 17:13:32 +08:00
|
|
|
|
2020-12-09 14:45:25 +08:00
|
|
|
losses = {'loss':total_loss, "score_loss":score_loss,\
|
|
|
|
"border_loss":border_loss, 'tvo_loss':tvo_loss, 'tco_loss':tco_loss}
|
2020-12-21 17:13:32 +08:00
|
|
|
return losses
|