Fix conceptually incorrect, non-breaking image shape tuple unpacking. - #582
Open
SSKlearns wants to merge 1 commit into
Open
Fix conceptually incorrect, non-breaking image shape tuple unpacking.#582SSKlearns wants to merge 1 commit into
SSKlearns wants to merge 1 commit into
Conversation
meta-cla
Bot
added
the
CLA Signed
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
In the current implementation, the code unpacks the tensor shape as: B, nc, w, h = x.shape
In PyTorch, the standard memory layout for 4D tensors is NCHW. By assigning the third dimension to W and the fourth to H, the spatial dimensions are logically inverted. For models trained on square images, this has zero impact on performance as H=W.
If this ViT class is used in future research for non-square inputs (e.g., 384×512), the positional embeddings will be calculated using incorrect spatial coordinates. This leads to a misalignment between the input patches and their corresponding positional encodings, potentially degrading model convergence and downstream performance.
I believe this error has propagated from DINO, and is not existent in DINOv3.