🆕 Add SegFormer Architecture

[Jiaqi-Lv]

This pull request adds SegFormer segmentation model, and refactors the EfficientUNetTissueMaskModel to use shared utility modules for common layers.

New architectures for semantic segmentation

SegFormer
Mix Transformer (Encoder)

Refactoring for code reuse

• Refactors EfficientUNetTissueMaskModel to use the shared Conv2dReLU and SegmentationHead implementations from architecture.utils, replacing local definitions and reducing code duplication. [1] [2] [3] [4] [5]

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.89%. Comparing base (14f59ab) to head (26ea885).

Additional details and impacted files

@@            Coverage Diff            @@
##           develop    #1085    +/-   ##
=========================================
  Coverage    99.88%   99.89%            
=========================================
  Files           86       88     +2     
  Lines        11670    11969   +299     
  Branches      1533     1550    +17     
=========================================
+ Hits         11657    11956   +299     
  Misses           7        7            
  Partials         6        6            

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR introduces a new transformer-based semantic segmentation architecture (SegFormer) backed by a Mix Transformer encoder, while refactoring existing segmentation architectures to reuse shared utility layers (e.g., Conv2dReLU, SegmentationHead, encoder utilities).

Changes:

• Add new SegFormer model implementation and a Mix Transformer (MiT) encoder backbone.
• Centralize shared encoder and convolution utilities in architecture.utils and remove duplicated implementations.
• Add/adjust unit tests for the new architecture and for the moved utility functions.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 13 comments.

Show a summary per file

File	Description
tiatoolbox/models/architecture/utils.py	Adds shared building blocks (Conv2dReLU, EncoderMixin, dilation/patching utilities) used across architectures.
tiatoolbox/models/architecture/timm_efficientnet.py	Refactors EfficientNet encoder to import shared encoder mixin utilities instead of local copies.
tiatoolbox/models/architecture/segformer.py	Adds SegFormer model + decoder and inference/pre/post-processing helpers.
tiatoolbox/models/architecture/mix_transformer.py	Adds Mix Transformer backbone/encoder implementation used by SegFormer.
tiatoolbox/models/architecture/grandqc.py	Replaces local Conv2dReLU with the shared implementation from architecture.utils.
tiatoolbox/models/architecture/efficientunet_tissue_mask_model.py	Refactors to use shared Conv2dReLU and SegmentationHead from architecture.utils.
tests/models/test_arch_utils.py	Adds tests for moved utilities (patch_first_conv, replace_strides_with_dilation).
tests/models/test_arch_timm_efficientnet.py	Removes tests that moved to test_arch_utils.py.
tests/models/test_arch_segformer.py	Adds unit tests covering SegFormer model, decoder behavior, and preprocessing/postprocessing/inference.
tests/models/test_arch_mix_transformer.py	Adds unit tests for Mix Transformer backbone + encoder behaviors.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review is ineligible. To be eligible to request a review, you need a paid Copilot license, or your organization must enable Copilot code review.

[shaneahmed]

The license is not compatible with TIAToolbox. Can we only include code which is BSD-3 compatible? Or clone the repo when it's requested with appropriate license displayed?

[shaneahmed]

Thanks @Jiaqi-Lv I am only concerned about the licence terms. These are not compatible with TIAToolbox licence. Otherwise the PR looks good.

[Jiaqi-Lv]

I'll look into using timm implementation of SegFormer, which has Apache License.