62 lines
2.2 KiB
Python
62 lines
2.2 KiB
Python
|
# copyright (c) 2022 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 .basic_loss import CELoss, DistanceLoss
|
||
|
|
|
||
|
|
|
||
|
|
class RFLLoss(nn.Layer):
|
||
|
|
def __init__(self, ignore_index=-100, **kwargs):
|
||
|
|
super().__init__()
|
||
|
|
|
||
|
|
self.cnt_loss = nn.MSELoss(**kwargs)
|
||
|
|
self.seq_loss = nn.CrossEntropyLoss(ignore_index=ignore_index)
|
||
|
|
|
||
|
|
def forward(self, predicts, batch):
|
||
|
|
|
||
|
|
self.total_loss = {}
|
||
|
|
total_loss = 0.0
|
||
|
|
# batch [image, label, length, cnt_label]
|
||
|
|
if predicts[0] is not None:
|
||
|
|
cnt_loss = self.cnt_loss(predicts[0],
|
||
|
|
paddle.cast(batch[3], paddle.float32))
|
||
|
|
self.total_loss['cnt_loss'] = cnt_loss
|
||
|
|
total_loss += cnt_loss
|
||
|
|
|
||
|
|
if predicts[1] is not None:
|
||
|
|
targets = batch[1].astype("int64")
|
||
|
|
label_lengths = batch[2].astype('int64')
|
||
|
|
batch_size, num_steps, num_classes = predicts[1].shape[0], predicts[
|
||
|
|
1].shape[1], predicts[1].shape[2]
|
||
|
|
assert len(targets.shape) == len(list(predicts[1].shape)) - 1, \
|
||
|
|
"The target's shape and inputs's shape is [N, d] and [N, num_steps]"
|
||
|
|
|
||
|
|
inputs = predicts[1][:, :-1, :]
|
||
|
|
targets = targets[:, 1:]
|
||
|
|
|
||
|
|
inputs = paddle.reshape(inputs, [-1, inputs.shape[-1]])
|
||
|
|
targets = paddle.reshape(targets, [-1])
|
||
|
|
seq_loss = self.seq_loss(inputs, targets)
|
||
|
|
self.total_loss['seq_loss'] = seq_loss
|
||
|
|
total_loss += seq_loss
|
||
|
|
|
||
|
|
self.total_loss['loss'] = total_loss
|
||
|
|
return self.total_loss
|