yujunjun/PaddleOCR
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2025-12-28 15:38:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix attenton loss for ce

Browse Source
This commit is contained in:
tink2123 2021-11-22 12:07:32 +08:00
parent 1bbf6e6a92
commit f26846ccd1
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ppocr/modeling/heads/rec_att_head.py
View File

@ -45,6 +45,7 @@ class AttentionHead(nn.Layer):
output_hiddens = []
if targets is not None:
print("target is not None")
for i in range(num_steps):
char_onehots = self._char_to_onehot(
targets[:, i], onehot_dim=self.num_classes)
@ -53,8 +54,8 @@ class AttentionHead(nn.Layer):
output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
output = paddle.concat(output_hiddens, axis=1)
probs = self.generator(output)
else:
print("target is None")
targets = paddle.zeros(shape=[batch_size], dtype="int32")
probs = None
char_onehots = None
@ -75,6 +76,7 @@ class AttentionHead(nn.Layer):
probs_step, axis=1)], axis=1)
next_input = probs_step.argmax(axis=1)
targets = next_input
if not self.training:
probs = paddle.nn.functional.softmax(probs, axis=2)
return probs