ModelListedExternal

YOLO26s RDD FRDC Distilled v2

A YOLO26s road-damage detector (4-class CRDDC: D00 longitudinal, D10 transverse, D20 alligator, D40 pothole) on the Unified Road Defect Dataset.

Sourcehuggingface.co/TamAko783/YOLO26s_RDD_FRDC_Distilled_v2/tree/448b62fb2e42d73f6c1fd209597d4e04cd465d55

KindOtherVersionRevision 448b62f

Model source

Kind: Other
File size: 19.4 MB
Version: vRevision 448b62f
Updated: Sep 19, 2026
Downloads: 118
Source: Hugging Face

Source description

YOLO26s · Distilled v2 (small, 2 teachers)

A YOLO26s road-damage detector (4-class CRDDC: D00 longitudinal, D10 transverse, D20 alligator, D40 pothole) on the Unified Road Defect Dataset.

Method: Semi-supervised distillation, two teachers (Co-DETR + RTMDet, WBF) on the larger YOLO26s student. Best overall model in the suite.

Metrics — RDD held-out validation (4,509 images, 11,470 boxes)

Class	mAP@50	mAP@50-95	Precision	Recall	F1
D00 Longitudinal	0.635	0.357	0.720	0.561	0.630
D10 Transverse	0.664	0.345	0.715	0.589	0.646
D20 Alligator	0.727	0.409	0.743	0.647	0.692
D40 Pothole	0.741	0.390	0.774	0.668	0.717
Overall	0.692	0.375	0.738	0.616	0.672

Full model comparison (same held-out val)

All five models in this study, evaluated identically (imgsz 640):

Model	Variant	Params	Distillation	mAP@50	mAP@50-95	F1
YOLO26n_RDD_Base	YOLO26n	2.4M	— (GT only)	0.635	0.334	0.621
YOLO26n_RDD_FRDC_Distilled	YOLO26n	2.4M	1 teacher (Co-DETR)	0.640	0.337	0.625
YOLO26n_RDD_FRDC_Distilled_v2	YOLO26n	2.4M	2 teachers (Co-DETR+RTMDet)	0.638	0.337	0.626
YOLO26s_RDD_Base	YOLO26s	9M	— (GT only)	0.687	0.372	0.665
➤ YOLO26s_RDD_FRDC_Distilled_v2 (this model)	YOLO26s	9M	2 teachers (Co-DETR+RTMDet)	0.692	0.375	0.672

Reading it:

Distillation helps — every distilled model beats its GT-only baseline.
Capacity helps most — the YOLO26s models (+~0.05 mAP@50) clearly outperform the nano ones on this val.
One vs two teachers is a near-tie at nano size; the two-teacher set's edge is small.
Cross-domain check (independent RDDC2024-ID, Indonesia, 8,901 imgs): distilled models generalized better than baselines, while the larger GT-only model generalized worse — evidence the distillation's added data improves robustness, not just in-domain fit.

RDD ground truth has known missing annotations, so absolute precision/recall are conservative for all models. The comparison is fair — every model uses the identical held-out val, never trained on.

Usage

from ultralytics import YOLO
model = YOLO("YOLO26s_RDD_FRDC_Distilled_v2.pt")
results = model("road.jpg")

Models in this suite

Datasets

Base: TamAko783/Unified_Road_Defect_Dataset
Distillation sets: v1 · v2 (two-teacher)

Credits

Datasets: RDD-2022 (Arya et al.) · UAV-PDD2023 · RoadDamageVision (Silva Zendron & Leithardt, CC BY 4.0).
Distillation teachers: Co-DETR Swin-L + RTMDet-x — FRDC (Wang Fangjun et al.), ORDDC'2024 winner.
Student: YOLO26s, AGPL-3.0 (Ultralytics).

Sources

1 source

Hugging FacePrimaryVersion 448b62fb2e42d73f6c1fd209597d4e04cd465d55

huggingface.co/TamAko783/YOLO26s_RDD_FRDC_Distilled_v2/tree/448b62fb2e42d73f6c1fd209597d4e04cd465d55

Verified Sep 9

Model artifacts

1 artifact

YOLO26s_RDD_FRDC_Distilled_v2.pt

pt · 19.4 MB · SHA-256 1e7ebe925286…b141 · Hugging Face

Download file

Source excerpts

2 excerpts

ReadmeSep 9huggingface.co/TamAko783/YOLO26s_RDD_FRDC_Distilled_v2

--- license: agpl-3.0 library_name: ultralytics pipeline_tag: object-detection tags: - yolo - yolo26 - road-damage - pothole-detection - crack-detection - knowledge-distillation model-index: - name: YOLO26s_RDD_FRDC_Distilled_v2 results: - task: {type: object-detection, name: Road Damage Detection} dataset: {name: Unified Road Defect Dataset (held-out val, 4509 imgs), type: TamAko783/Unified_Road_Defect_Dataset} metrics: - {type: mAP50, value: 0.692, name: mAP@50} - {type: mAP50-95, value: 0.375, name: mAP@50-95} - {type: F1, value: 0.672, name: F1} --- # YOLO26s · Distilled v2 (small, 2 teachers) A **YOLO26s** road-damage detector (4-class CRDDC: D00 longitudinal, D10 transverse, D20 alligator, D40 pothole) on the [Unified Road Defect Dataset](https://huggingface.co/datasets/TamAko783/Unified_Road_Defect_Dataset). **Method:** **Semi-supervised distillation, two teachers** (Co-DETR + RTMDet, WBF) on the larger YOLO26s student. Best overall model in the suite. ## Metrics — RDD held-out validation (4,509 images, 11,470 boxes) | Class | mAP@50 | mAP@50-95 | Precision | Recall | F1 | |---|---:|---:|---:|---:|---:| | D00 Longitudinal | 0.635 | 0.357 | 0.720 | 0.561 | 0.630 | | D10 Transverse | 0.664 | 0.345 | 0.715 | 0.589 | 0.646 | | D20 Alligator | 0.727 | 0.409 | 0.743 | 0.647 | 0.692 | | D40 Pothole | 0.741 | 0.390 | 0.774 | 0.668 | 0.717 | | **Overall** | **0.692** | **0.375** | **0.738** | **0.616** | **0.672** | ## Full model comparison (same held-out val) All five models in this study, evaluated identically (imgsz 640): | Model | Variant | Params | Distillation | mAP@50 | mAP@50-95 | F1 | |---|---|---:|---|---:|---:|---:| | [YOLO26n_RDD_Base](https://huggingface.co/TamAko783/YOLO26n_RDD_Base) | YOLO26n | 2.4M | — (GT only) | 0.635 | 0.334 | 0.621 | | [YOLO26n_RDD_FRDC_Distilled](https://huggingface.co/TamAko783/YOLO26n_RDD_FRDC_Distilled) | YOLO26n | 2.4M | 1 teacher (Co-DETR) | 0.640 | 0.337 | 0.625 | | [YOLO26n_RDD_FRDC_Distilled_v2](https://huggingface.co/TamAko783/YOLO26n_RDD_FRDC_Distilled_v2) | YOLO26n | 2.4M | 2 teachers (Co-DETR+RTMDet) | 0.638 | 0.337 | 0.626 | | [YOLO26s_RDD_Base](https://huggingface.co/TamAko783/YOLO26s_RDD_Base) | YOLO26s | 9M | — (GT only) | 0.687 | 0.372 | 0.665 | | **➤ YOLO26s_RDD_FRDC_Distilled_v2** (this model) | YOLO26s | 9M | 2 teachers (Co-DETR+RTMDet) | **0.692** | **0.375** | **0.672** | **Reading it:...

NoteSep 9huggingface.co/TamAko783/YOLO26s_RDD_FRDC_Distilled_v2

Source context: 118 downloads · 0 likes · Pipeline object-detection · Library ultralytics · Repo TamAko783/YOLO26s_RDD_FRDC_Distilled_v2