Add EdgeCrafter Model

[kprokofi]

Summary

How to test

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	4.413	0.003
xpu	4.432	0.003
cuda	4.666	0.003

[github-actions]

📊 Test coverage report

Metric	Coverage
Lines	60.0%
Functions	56.7%
Branches	51.5%
Statements	59.6%

[github-actions]

🐳 Docker image sizes

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

[codecov-commenter]

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

Codecov Report

❌ Patch coverage is 29.26009% with 631 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...end/lightning/models/detection/heads/ec_decoder.py	11.64%	311 Missing ⚠️
...ckend/lightning/models/common/edgecrafter_mixin.py	33.33%	100 Missing ⚠️
...kend/lightning/models/detection/backbones/ecvit.py	24.39%	93 Missing ⚠️
...nstance_segmentation/heads/ec_segmentation_head.py	26.78%	41 Missing ⚠️
...ightning/models/detection/detectors/edgecrafter.py	23.80%	32 Missing ⚠️
...ckend/lightning/models/detection/losses/ec_loss.py	34.28%	23 Missing ⚠️
.../backend/lightning/models/detection/edgecrafter.py	62.16%	14 Missing ⚠️
...g/models/instance_segmentation/edgecrafter_inst.py	67.50%	13 Missing ⚠️
...models/common/utils/assigners/hungarian_matcher.py	0.00%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Copilot]

Pull request overview

This PR introduces the EdgeCrafter (ECDet) object detection model family into the GetiTune library and exposes the new architectures in the backend via supported-model manifests and config conversion mappings.

Changes:

• Added EdgeCrafter detection recipes and backend manifests for S/M/L/X variants.
• Implemented EdgeCrafter model components in the library (backbone, decoder/head, loss, detector wrapper, and shared mixin) and wired them into existing registries/config tables.
• Updated backend model-architectures endpoint tests to account for the additional detection manifests.

Reviewed changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
library/src/getitune/recipe/detection/edgecrafter_s.yaml	Adds ECDet-S training recipe configuration.
library/src/getitune/recipe/detection/edgecrafter_m.yaml	Adds ECDet-M training recipe configuration.
library/src/getitune/recipe/detection/edgecrafter_l.yaml	Adds ECDet-L training recipe configuration.
library/src/getitune/recipe/detection/edgecrafter_x.yaml	Adds ECDet-X training recipe configuration.
library/src/getitune/backend/lightning/models/detection/edgecrafter.py	Adds Lightning detection model wrapper for EdgeCrafter.
library/src/getitune/backend/lightning/models/instance_segmentation/edgecrafter_inst.py	Adds Lightning instance-segmentation wrapper for EdgeCrafter (ECSeg weights).
library/src/getitune/backend/lightning/models/instance_segmentation/__init__.py	Exports EdgeCrafterInst from the instance segmentation models package.
library/src/getitune/backend/lightning/models/detection/necks/dfine_hybrid_encoder.py	Registers EdgeCrafter variant configs for the hybrid encoder.
library/src/getitune/backend/lightning/models/detection/losses/ec_loss.py	Adds ECCriterion with optional mask losses for instance segmentation.
library/src/getitune/backend/lightning/models/detection/losses/__init__.py	Exposes ECCriterion from the detection losses package.
library/src/getitune/backend/lightning/models/detection/heads/ec_decoder.py	Adds ECTransformer decoder and segmentation head implementation.
library/src/getitune/backend/lightning/models/detection/detectors/edgecrafter.py	Adds DETR-style detector wrapper integrating the ECTransformer and optional masks.
library/src/getitune/backend/lightning/models/detection/detectors/__init__.py	Exposes ECDETRDetector from the detectors package.
library/src/getitune/backend/lightning/models/detection/backbones/ecvit.py	Adds ECViTAdapter backbone implementation for EdgeCrafter.
library/src/getitune/backend/lightning/models/detection/__init__.py	Exposes EdgeCrafter from the detection models package.
library/src/getitune/backend/lightning/models/common/utils/assigners/hungarian_matcher.py	Handles missing pred_masks safely in mask cost computation.
library/src/getitune/backend/lightning/models/common/edgecrafter_mixin.py	Adds shared model construction, IO adaptation, and optimizer config for EdgeCrafter.
application/backend/app/supported_models/manifests/detection/edgecrafter_s.yaml	Adds backend manifest for ECDet-S.
application/backend/app/supported_models/manifests/detection/edgecrafter_m.yaml	Adds backend manifest for ECDet-M.
application/backend/app/supported_models/manifests/detection/edgecrafter_l.yaml	Adds backend manifest for ECDet-L.
application/backend/app/supported_models/manifests/detection/edgecrafter_x.yaml	Adds backend manifest for ECDet-X.
application/backend/app/execution/common/geti_config_converter.py	Registers EdgeCrafter template IDs to map to GetiTune recipes.
application/backend/tests/unit/api/routers/test_model_architectures.py	Updates expected detection model counts for manifests endpoint.

[kprokofi]

[AlbertvanHouten]

It would be good to include one of the new models in the benchmark_manifest.

[kprokofi]

detection

Model	Dataset	Scenario	Val f1-score ↑	Val map ↑	Val map_50 ↑	Val map_75 ↑	Val map_large ↑	Val map_medium ↑	Val map_per_class ↑	Val map_small ↑	Val mar_1 ↑	Val mar_10 ↑	Val mar_100 ↑	Val mar_100_per_class ↑	Val mar_large ↑	Val mar_medium ↑	Val mar_small ↑	Train Time ↓	GPU Mem ↓	Test Latency ↓	Status
deimv2_l	axial_mri	default	0.6522	0.7292	0.9505	0.8939	0.8465	0.7063	-1.0000	0.9000	0.6638	0.8064	0.8638	-1.0000	0.9125	0.8600	0.9000	5m 38s	6.1 GB	0.1s	🆕
deimv2_l	bccd	default	0.8237	0.6502	0.9145	0.7522	0.7476	0.5272	-1.0000	0.1965	0.4220	0.6860	0.7870	-1.0000	0.8439	0.7299	0.3333	9m 26s	6.1 GB	0.2s	🆕
deimv2_l	cable_damage	default	0.9220	0.4697	0.9351	0.4330	0.5762	0.4551	-1.0000	0.1836	0.5138	0.5900	0.7015	-1.0000	0.7906	0.7079	0.3800	17m 43s	6.1 GB	0.1s	🆕
deimv2_l	vehicles	default	0.6858	0.2969	0.4281	0.3354	0.4527	0.3110	-1.0000	0.2689	0.3641	0.6956	0.7259	-1.0000	0.8200	0.7282	0.5665	106m 21s	6.3 GB	0.1s	🆕
edgecrafter_l	axial_mri	default	0.8913	0.7298	0.9842	0.9177	0.7582	0.7382	-1.0000	0.8000	0.6574	0.8213	0.8553	-1.0000	0.8938	0.8533	0.9000	7m 20s	3.2 GB	0.1s	🆕
edgecrafter_l	bccd	default	0.8340	0.6572	0.9189	0.7769	0.7534	0.5461	-1.0000	0.2139	0.4217	0.6859	0.7775	-1.0000	0.8534	0.7204	0.3278	16m 11s	3.3 GB	0.2s	🆕
edgecrafter_l	cable_damage	default	0.9267	0.4427	0.9142	0.3882	0.5378	0.4339	-1.0000	0.2663	0.4918	0.5904	0.6349	-1.0000	0.6534	0.6559	0.3800	39m 58s	3.2 GB	0.1s	🆕
edgecrafter_l	vehicles	default	0.6901	0.3047	0.4420	0.3384	0.4371	0.3227	-1.0000	0.2776	0.3697	0.6982	0.7235	-1.0000	0.8224	0.7258	0.5895	146m 27s	3.4 GB	0.1s	🆕
edgecrafter_m	axial_mri	default	0.7677	0.6874	0.9576	0.8318	0.7329	0.6848	-1.0000	1.0000	0.6596	0.8128	0.8638	-1.0000	0.8875	0.8700	1.0000	4m 35s	4.6 GB	0.1s	🆕
edgecrafter_m	bccd	default	0.8312	0.6489	0.9136	0.7687	0.7508	0.5440	-1.0000	0.1879	0.4211	0.6855	0.7872	-1.0000	0.8483	0.7306	0.3111	10m 51s	4.7 GB	0.2s	🆕
edgecrafter_m	cable_damage	default	0.9184	0.4671	0.9291	0.3724	0.5689	0.4533	-1.0000	0.2455	0.4964	0.5686	0.6630	-1.0000	0.7680	0.6693	0.3800	16m 13s	4.6 GB	0.1s	🆕
edgecrafter_m	vehicles	default	0.6809	0.2829	0.4180	0.3164	0.4213	0.2954	-1.0000	0.2240	0.3532	0.6647	0.7149	-1.0000	0.8163	0.7127	0.5658	47m 40s	4.9 GB	0.1s	🆕
edgecrafter_s	axial_mri	default	0.8454	0.6790	0.9590	0.8284	0.7289	0.6863	-1.0000	0.7333	0.6404	0.8064	0.8915	-1.0000	0.8938	0.8933	0.9000	4m 43s	3.4 GB	0.1s	🆕
edgecrafter_s	bccd	default	0.8282	0.6446	0.9075	0.7590	0.7507	0.5226	-1.0000	0.2500	0.4202	0.6849	0.7716	-1.0000	0.8526	0.7078	0.3111	12m 15s	3.4 GB	0.2s	🆕
edgecrafter_s	cable_damage	default	0.9178	0.4445	0.9314	0.3856	0.5662	0.4321	-1.0000	0.2137	0.4740	0.5618	0.6479	-1.0000	0.7602	0.6444	0.3800	17m 4s	3.4 GB	0.1s	🆕
edgecrafter_s	vehicles	default	0.6475	0.2199	0.3534	0.2389	0.3915	0.2473	-1.0000	0.1943	0.3060	0.6064	0.6363	-1.0000	0.7885	0.6358	0.4443	43m 1s	3.7 GB	0.1s	🆕
edgecrafter_x	axial_mri	default	0.7387	0.6850	0.9673	0.8974	0.7491	0.7132	-1.0000	1.0000	0.6851	0.8362	0.8702	-1.0000	0.8938	0.8633	1.0000	11m 47s	4.1 GB	0.2s	🆕
edgecrafter_x	bccd	default	0.8117	0.6426	0.9064	0.7621	0.7446	0.5282	-1.0000	0.2095	0.4143	0.6853	0.7734	-1.0000	0.8575	0.7112	0.2944	8m 3s	4.1 GB	0.2s	🆕
edgecrafter_x	cable_damage	default	0.9103	0.4450	0.9239	0.3873	0.5717	0.4304	-1.0000	0.1703	0.4741	0.5774	0.6927	-1.0000	0.7947	0.6981	0.4400	23m 58s	4.0 GB	0.1s	🆕
edgecrafter_x	vehicles	default	0.7000	0.3025	0.4472	0.3391	0.4472	0.3239	-1.0000	0.2878	0.3631	0.6931	0.7234	-1.0000	0.8207	0.7208	0.5835	145m 28s	4.2 GB	0.1s	🆕
rfdetr_large	axial_mri	default	0.7350	0.7501	0.9994	0.9457	0.8171	0.7893	-1.0000	0.8000	0.6596	0.8489	0.8489	-1.0000	0.8813	0.8400	0.9000	8m 45s	5.3 GB	0.2s	🆕
rfdetr_large	bccd	default	0.8341	0.6543	0.9212	0.7489	0.7565	0.5326	-1.0000	0.2209	0.4179	0.6747	0.7581	-1.0000	0.8385	0.7001	0.2722	10m 50s	5.3 GB	0.2s	🆕
rfdetr_large	cable_damage	default	0.9524	0.4658	0.9546	0.4138	0.5845	0.4501	-1.0000	0.2142	0.4991	0.5677	0.5907	-1.0000	0.6977	0.5860	0.3000	41m 1s	5.3 GB	0.1s	🆕
rfdetr_large	vehicles	default	0.6865	0.2976	0.4467	0.3343	0.4354	0.3129	-1.0000	0.2483	0.3448	0.6709	0.7061	-1.0000	0.8065	0.7098	0.5537	89m 43s	5.5 GB	0.1s	🆕

[kprokofi]

@itallix I applied all clean up suggestions in bbb56d1

Please, review once again

[kprokofi]

edge_crafter_l metrics on WGISD after clean up:

So, all good here

[Copilot]

Pull request overview

Copilot reviewed 30 out of 30 changed files in this pull request and generated 1 comment.