Skip to content

Fix: repeat per sample for SAM image embeddings #25074

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

Merged
merged 1 commit into from
Jul 25, 2023
Merged

Fix: repeat per sample for SAM image embeddings #25074

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 2 additions & 2 deletions src/transformers/models/sam/modeling_sam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -507,8 +507,8 @@ def forward(

# Expand per-image data in batch direction to be per-point
image_embeddings = image_embeddings + dense_prompt_embeddings
image_embeddings = image_embeddings.repeat(point_batch_size, 1, 1, 1)
image_positional_embeddings = image_positional_embeddings.repeat(point_batch_size, 1, 1, 1)
image_embeddings = image_embeddings.repeat_interleave(point_batch_size, 0)
image_positional_embeddings = image_positional_embeddings.repeat_interleave(point_batch_size, 0)

# Run the transformer, image_positional_embedding are consumed
point_embedding, image_embeddings, attentions = self.transformer(
Expand Down
4 changes: 2 additions & 2 deletions src/transformers/models/sam/modeling_tf_sam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,8 +517,8 @@ def call(
point_embeddings = tf.cast(tokens, self.iou_token.dtype)

image_embeddings = image_embeddings + dense_prompt_embeddings
image_embeddings = tf.tile(image_embeddings, [point_batch_size, 1, 1, 1])
image_positional_embeddings = tf.tile(image_positional_embeddings, [point_batch_size, 1, 1, 1])
image_embeddings = tf.repeat(image_embeddings, point_batch_size, axis=0)
image_positional_embeddings = tf.repeat(image_positional_embeddings, point_batch_size, axis=0)

point_embedding, image_embeddings, attentions = self.transformer(
point_embeddings=point_embeddings,
Expand Down