Update tensor.rs to Handle Trailing Unit-Dimensions in Image Tensors #1916
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Previously, tensor.rs only displayed tensors with shapes of [H, W, 1|3|4] as images, this update addresses the issue by modifying the `image_height_width_channels()`, `is_shaped_like_an_image()`, and `is_vector()` methods to ignore trailing unit-dimensions when checking the shape of the tensor - This change should improve the user experience by allowing a wider range of image tensor shapes to be displayed as images in tensor.rs
emilk
requested changes
Apr 19, 2023
| @@ -424,20 +435,33 @@ impl Tensor { | |||
| } | |||
|
|
|||
| pub fn is_shaped_like_an_image(&self) -> bool { | |||
There was a problem hiding this comment.
we can simplify this quite a bit:
Suggested change
| pub fn is_shaped_like_an_image(&self) -> bool { | |
| pub fn is_shaped_like_an_image(&self) -> bool { | |
| self.image_height_width_channels().is_some() | |
| } |
| } | ||
|
|
||
| #[inline] | ||
| pub fn is_vector(&self) -> bool { | ||
| let shape = &self.shape; | ||
| shape.len() == 1 || { shape.len() == 2 && (shape[0].size == 1 || shape[1].size == 1) } | ||
| let mut last_dim = shape.last().unwrap().size; |
There was a problem hiding this comment.
this will panic if shape is empty (for "scalar tensors")
Comment on lines
+411
to
+420
| let mut shape = self.shape.clone(); | ||
|
|
||
| // Remove trailing unit dimensions | ||
| while let Some(&d) = shape.last() { | ||
| if d.size == 1 { | ||
| shape.pop(); | ||
| } else { | ||
| break; | ||
| } | ||
| } |
There was a problem hiding this comment.
I suggest encapsulating this in a helper function, e.g:
/// Returns the shape, but ignoring trailing dimensions of length 1.
///
/// For instance, a shape of `[640, 480, 1, 1]` would be returned as `[640, 480]`.
fn short_shape(&self) -> Vec<TensorDimension> {
…
}this could then be used by image_height_width_channels as well as is_vector
|
Thanks for the PR! It would be great if you added a simple test of this to |
emilk
added
enhancement
|
@tauseefmohammed2 you're still on that? :) |
|
Took the task over in #2075 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What
Previously, tensor.rs only displayed tensors with shapes of [H, W, 1|3|4] as images, this update addresses the issue by modifying the
image_height_width_channels(),is_shaped_like_an_image(), andis_vector()methods to ignore trailing unit-dimensions when checking the shape of the tensorThis change should improve the user experience by allowing a wider range of image tensor shapes to be displayed as images in tensor.rs
Resolves Issue: Show tensors shaped
[H, W, 1, 1]as images #1871Checklist