Support tiling

[A-Artemis]

Summary

Added TilingDatasetFactory for wrapping the dataset subsets if tiling is enabled.
Created tiling config yamls for algorithms which were missing it.

Checklist

• The PR title and description are clear and descriptive
• I have manually tested the changes
• All changes are covered by automated tests
• All related issues are linked to this PR (if applicable)
• Documentation has been updated (if applicable)

[github-actions]

⏱️ Backend import time — app.main

Accelerator	Cumulative (s)	Self (s)
cpu	4.628	0.003
xpu	4.476	0.003
cuda	4.726	0.003

[github-actions]

📊 Test coverage report

Metric	Coverage
Lines	59.8%
Functions	56.5%
Branches	51.1%
Statements	59.4%

[github-actions]

🐳 Docker image sizes

Device	Size
cpu	4002.9 MB (3.91 GB)
cuda	10198.1 MB (9.96 GB)
xpu	10791.5 MB (10.54 GB)

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 72.13115% with 17 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...e/backend/openvino/models/instance_segmentation.py	23.07%	10 Missing ⚠️
...brary/src/getitune/backend/openvino/models/base.py	55.55%	4 Missing ⚠️
.../src/getitune/backend/openvino/models/detection.py	92.30%	1 Missing ⚠️
library/src/getitune/config/data.py	50.00%	1 Missing ⚠️
library/src/getitune/data/augmentation/pipeline.py	88.88%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Copilot]

Pull request overview

Enables end-to-end tiling support across the GetiTune library and the application backend, covering dataset wrapping, GPU augmentation handling for tile batches, and tile-aware OpenVINO inference/merging, plus adding missing tiling recipes.

Changes:

• Add tile-batch compatible GPU augmentation path (image-only transforms) and corresponding unit tests.
• Add tile-aware OpenVINO detection/instance-segmentation inference (forward over tiles + merge) and dtype regression coverage.
• Add/extend tiling recipes and update backend config conversion/training pipeline to enable tiling while skipping incompatible augmentations.

Reviewed changes

Copilot reviewed 49 out of 49 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
library/tests/unit/data/augmentation/test_pipeline.py	Adds regression tests for GPU augmentation on tile batches.
library/tests/unit/backend/openvino/models/test_detection.py	Adds tiling-focused tests and a float64→float32 score dtype regression test.
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_xlarge_tile.yaml	New tiling recipe for RFDETR instance segmentation (xlarge).
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_small_tile.yaml	New tiling recipe for RFDETR instance segmentation (small).
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_nano_tile.yaml	New tiling recipe for RFDETR instance segmentation (nano).
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_medium_tile.yaml	New tiling recipe for RFDETR instance segmentation (medium).
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_large_tile.yaml	New tiling recipe for RFDETR instance segmentation (large).
library/src/getitune/recipe/instance_segmentation/rfdetr_seg_2xlarge_tile.yaml	New tiling recipe for RFDETR instance segmentation (2xlarge).
library/src/getitune/recipe/detection/rfdetr_small_tile.yaml	New tiling recipe for RFDETR detection (small).
library/src/getitune/recipe/detection/rfdetr_nano_tile.yaml	New tiling recipe for RFDETR detection (nano).
library/src/getitune/recipe/detection/rfdetr_medium_tile.yaml	New tiling recipe for RFDETR detection (medium).
library/src/getitune/recipe/detection/rfdetr_large_tile.yaml	New tiling recipe for RFDETR detection (large).
library/src/getitune/recipe/detection/deimv2_s_tile.yaml	New tiling recipe for DEIMv2 (small).
library/src/getitune/recipe/detection/deimv2_m_tile.yaml	New tiling recipe for DEIMv2 (medium).
library/src/getitune/recipe/detection/deimv2_l_tile.yaml	New tiling recipe for DEIMv2 (large).
library/src/getitune/recipe/detection/deim_dfine_x_tile.yaml	New tiling recipe for DEIM D-FINE (x).
library/src/getitune/recipe/detection/deim_dfine_m_tile.yaml	New tiling recipe for DEIM D-FINE (m).
library/src/getitune/recipe/detection/deim_dfine_l_tile.yaml	New tiling recipe for DEIM D-FINE (l).
library/src/getitune/data/augmentation/pipeline.py	Adds apply_image_only() to support tile-batch normalization/intensity transforms without annotations.
library/src/getitune/config/data.py	Adds a custom TileConfig.__repr__ for clearer logging/debugging.
library/src/getitune/backend/openvino/models/instance_segmentation.py	Adds tile-aware forward/merge path and fixes score dtype for NMS compatibility.
library/src/getitune/backend/openvino/models/detection.py	Adds tile-aware forward/merge path and fixes score dtype for NMS compatibility.
library/src/getitune/backend/openvino/models/base.py	Routes TileBatch inputs to forward_tiles() in test_step / predict_step; adds default tile_config.
library/src/getitune/backend/openvino/engine.py	Propagates datamodule.tile_config into OV model during test().
library/src/getitune/backend/lightning/callbacks/gpu_augmentation.py	Applies GPU augmentations safely to tile batches (per-tile, image-only).
library/benchmark_manifest.yaml	Adds tiling variants to detection/instance-seg benchmarks and reorganizes model lists.
application/backend/tests/unit/execution/training/test_getitune_trainer.py	Updates mocks to provide working __len__ for dataset size logging.
application/backend/tests/unit/execution/common/test_geti_config_converter.py	Adds tests for DEIM↔tiling incompatibility and ensures mosaic/mixup are skipped under tiling.
application/backend/tests/unit/api/schemas/test_training_configuration_view.py	Updates augmentation descriptions to note tiling incompatibilities.
application/backend/tests/integration/services/test_training_configuration_service.py	Adjusts tiling capability test coverage to use a known tiling-unsupported architecture.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_xl.yaml	Removes explicit capabilities.tiling: false to allow tiling per defaults.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_s.yaml	Same as above for RFDETR-S instance segmentation.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_n.yaml	Same as above for RFDETR-N instance segmentation.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_m.yaml	Same as above for RFDETR-M instance segmentation.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_l.yaml	Same as above for RFDETR-L instance segmentation.
application/backend/app/supported_models/manifests/instance_segmentation/rfdetr_2xl.yaml	Same as above for RFDETR-2XL instance segmentation.
application/backend/app/supported_models/manifests/detection/rfdetr_s.yaml	Removes explicit capabilities.tiling: false to allow tiling per defaults.
application/backend/app/supported_models/manifests/detection/rfdetr_n.yaml	Same as above for RFDETR-N detection.
application/backend/app/supported_models/manifests/detection/rfdetr_m.yaml	Same as above for RFDETR-M detection.
application/backend/app/supported_models/manifests/detection/rfdetr_l.yaml	Same as above for RFDETR-L detection.
application/backend/app/supported_models/manifests/detection/dinov3_detr_s.yaml	Removes explicit capabilities.tiling: false to allow tiling per defaults.
application/backend/app/supported_models/manifests/detection/dinov3_detr_m.yaml	Same as above for DINOv3-DETR-M.
application/backend/app/supported_models/manifests/detection/dinov3_detr_l.yaml	Same as above for DINOv3-DETR-L.
application/backend/app/supported_models/manifests/detection/dfine_x.yaml	Removes explicit capabilities.tiling: false to allow tiling per defaults.
application/backend/app/supported_models/manifests/detection/dfine_m.yaml	Same as above for D-FINE-M.
application/backend/app/supported_models/manifests/detection/dfine_l.yaml	Same as above for D-FINE-L.
application/backend/app/models/training_configuration/augmentation.py	Updates DEIM/mosaic/mixup descriptions to document tiling incompatibilities.
application/backend/app/execution/training/getitune_trainer.py	Wraps datasets with tiling factory when enabled and propagates resolved tile_config into datamodule/model.
application/backend/app/execution/common/geti_config_converter.py	Skips tiling-incompatible augmentations and forces DEIM off when tiling is enabled.