Hugging Face's logo

TamAko783
/
YOLO26n_RDD_FRDC_Distilled

Object Detection
ultralytics
yolo
yolo26
road-damage
pothole-detection
crack-detection
knowledge-distillation
Eval Results (legacy)
Model card Files
xet
 Community
YOLO26n_RDD_FRDC_Distilled / README.md
TamAko783's picture
TamAko783
Update card: RDD-val per-class metrics + full 5-model comparison + links
ce0bd64 verified
preview code
|
Raw
History Blame Contribute Delete
4.42 kB
metadata 
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: YOLO26n_RDD_FRDC_Distilled
    results:
      - task:
          type: object-detection
          name: Road Damage Detection
        dataset:
          name: Unified Road Defect Dataset (held-out val
          4509 imgs): null
          type: TamAko783/Unified_Road_Defect_Dataset
        metrics:
          - type: mAP50
            value: 0.64
            name: mAP@50
          - type: mAP50-95
            value: 0.337
            name: mAP@50-95
          - type: F1
            value: 0.625
            name: F1

YOLO26n · Distilled v1 (1 teacher: Co-DETR)

A YOLO26n road-damage detector (4-class CRDDC: D00 longitudinal, D10 transverse, D20 alligator, D40 pothole) on the Unified Road Defect Dataset.

Method: Semi-supervised distillation, single teacher. Co-DETR (Swin-L) pseudo-labeled the unlabeled RDD-2022 test set; student trained on GT + pseudo-labels.

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

Class mAP@50 mAP@50-95 Precision Recall F1
D00 Longitudinal 0.590 0.324 0.675 0.520 0.588
D10 Transverse 0.594 0.299 0.683 0.504 0.580
D20 Alligator 0.687 0.375 0.713 0.617 0.662
D40 Pothole 0.688 0.347 0.737 0.614 0.670
Overall 0.640 0.337 0.702 0.564 0.625

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 (this model) 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 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("YOLO26n_RDD_FRDC_Distilled.pt")
results = model("road.jpg")

Models in this suite

Datasets

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: YOLO26n, AGPL-3.0 (Ultralytics).