Skip to content
New issue

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

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

Already on GitHub? # to your account

Possible indentation error #2350

Closed
ariG23498 opened this issue Feb 15, 2024 · 4 comments · Fixed by #2360
Closed

Possible indentation error #2350

ariG23498 opened this issue Feb 15, 2024 · 4 comments · Fixed by #2360
Assignees
Labels
type:Bug Something isn't working

Comments

@ariG23498
Copy link
Contributor

With the following code it seems that the first roi would return the regions.

If this is correct, I would send in a PR for this fix.

@ariG23498
Copy link
Contributor Author

I wanted to report another issue in this thread.

Using the

feature_map = tf.random.normal([2, 16, 16, 512])
roi_pooler = ROIPooler(
    bounding_box_format="yxyx", target_size=[7, 7], image_shape=[224, 224, 3]
)
rois = tf.constant(
    [
        [[15.0, 30.0, 25.0, 45.0], [16.0, 31.0, 26.0, 46.0]],
        [[22.0, 1.0, 30.0, 32.0], [23.0, 2.0, 31.0, 33.0]],
    ],
)
print(f"{feature_map.shape=}")
print(f"{rois.shape=}")
pooled_feature_map = roi_pooler(feature_map, rois)
print(f"{pooled_feature_map.shape=}")

# feature_map.shape=TensorShape([2, 16, 16, 512])
# rois.shape=TensorShape([2, 2, 4])
# pooled_feature_map.shape=TensorShape([2, 7, 7, 512])

Notice how the pooled_feature_map.shape does not conside the number of rois.

GitHub Gist

CC: @divyashreepathihalli

This issue directly affects the #2295

@divyashreepathihalli
Copy link
Collaborator

Looks like a bug, do you have a fix PR?

@ariG23498
Copy link
Contributor Author

Unfortunately I do not.

@divyashreepathihalli
Copy link
Collaborator

Fixed.

# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
type:Bug Something isn't working
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants