Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Fix] PARSeq tensorflow fixes #1228

Merged
merged 7 commits into from
Jun 23, 2023
+79 −82
Merged

[Fix] PARSeq tensorflow fixes #1228

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
58402de
up
felixdittrich92 Jun 21, 2023
c9ed5bd
update
felixdittrich92 Jun 22, 2023
0b47c81
fix run
felixdittrich92 Jun 22, 2023
241b95a
update
felixdittrich92 Jun 22, 2023
e9aac29
update decoding
felixdittrich92 Jun 22, 2023
1121249
fix mask
felixdittrich92 Jun 22, 2023
c19eb9b
update comment
felixdittrich92 Jun 23, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
update decoding
  • Loading branch information
@felixdittrich92
felixdittrich92 committed Jun 23, 2023
commit e9aac295c11ec2864765bd57c84f6e45ccaf378d
9 changes: 6 additions & 3 deletions doctr/models/recognition/parseq/tensorflow.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -302,12 +302,15 @@ def decode_autoregressive(self, features: tf.Tensor, max_len: Optional[int] = No

sos = tf.fill((tf.shape(features)[0], 1), self.vocab_size + 1)
ys = tf.concat([sos, tf.cast(tf.argmax(logits[:, :-1], axis=-1), dtype=tf.int32)], axis=1)
# Create padding mask for refined target input maskes all behind EOS token as False
# Create padding mask for refined target input sequence
# (N, 1, 1, max_length)
target_pad_mask = tf.cast(
tf.expand_dims(tf.expand_dims(tf.not_equal(ys, self.vocab_size), axis=1), axis=1), tf.bool
tf.math.cumsum(tf.cast(tf.equal(ys, self.vocab_size), dtype=tf.int32), axis=-1, reverse=True) > 0,
dtype=tf.bool,
)
mask = tf.math.logical_or(target_pad_mask, query_mask[:, : ys.shape[1]])
target_pad_mask = target_pad_mask[:, tf.newaxis, tf.newaxis, :]

mask = tf.math.logical_and(target_pad_mask, query_mask[:, : ys.shape[1]])
logits = self.head(self.decode(ys, features, mask, target_query=pos_queries, **kwargs), **kwargs)

return logits # (N, max_length, vocab_size + 1)
Expand Down