You can deploy this model and start earning money today!

SaurabhKrGupta7788/
UVH-26

Name: UVH-26
Availability: InStock

Views

No views yet

ArXiv |

Dataset

Models trained on UVH-26 deliver up to 31.5% higher mAP than COCO-pretrained baselines, demonstrating significant gains in real-world performance for Indian traffic scenarios.

UVH-26 Vehicle Detection Models — AIM@IISc

High-quality object detection models built for Indian road traffic — where vehicle appearance, traffic density, and scene complexity differ significantly from Western datasets like COCO.

These models are trained on the UVH-26 dataset, featuring:

14 road-relevant vehicle categories
Real urban environments across India
Diverse viewpoints, lighting, occlusion & density variations
Multi-user labeled data with consensus filtering (MV / ST variants)

We currently release six SOTA detector variants trained on the dataset:

Model Family	Sizes	Strengths
YOLOv11	S, X	Fast + lightweight deployment
RT-DETRv2	S, X	High-accuracy, transformer-based real-time detection
DAMO-YOLO	T, L	Efficient architectures, strong large-scale performance

Designed for Indian mobility — adaptable to real city surveillance, roadside cameras, safety monitoring, and ITS applications.

Model Dataset -> https://huggingface.co/datasets/iisc-aim/UVH-26

Attribution

More technical details about the dataset and models are available in our Technical Report available on arXiv. If you use these datasets or models, kindly cite the following: The Urban Vision Hackathon Dataset and Models: Towards Image Annotations and Accurate Vision Models for Indian Traffic, Preliminary Dataset Release, UVH-26-v1.0, Akash Sharma, Chinmay Mhatre, Sankalp Gawali, Ruthvik Bokkasam, Brij Kishore, Vishwajeet Pattanaik, Tarun Rambha, Abdul R. Pinjari, Vijay Kovvali, Anirban Chakraborty, Punit Rathore, Raghu Krishnapuram and Yogesh Simmhan, Technical Report, Indian Institute of Science, arXiv:2511.02563, Nov, 2025.

bibtex

@techreport{sharma2025uvh26,
  title        = {Towards Image Annotations and Accurate Vision Models for Indian Traffic, Preliminary Dataset Release, UVH-26-v1.0},
  author       = {Akash Sharma and Chinmay Mhatre and Sankalp Gawali and Ruthvik Bokkasam and Brij Kishore and Vishwajeet Pattanaik and Tarun Rambha and Abdul R. Pinjari and Vijay Kovvali and Anirban Chakraborty and Punit Rathore and Raghu Krishnapuram and Yogesh Simmhan},
  institution  = {Indian Institute of Science},
  type         = {Technical Report},
  number       = {arXiv:2511.02563},
  year         = {2025},
  month        = {November},
  doi          = {10.48550/arXiv.2511.02563}
}

Repository Structure

LICENSE – Apache-2.0 license
uvh_classes.txt – 14 object classes (one per line)
configs/ – Model configuration files
- yolo11_s.yaml
- yolo11_x.yaml
- rtdetr_s.yaml
- rtdetr_x.yaml
- damo_yolo_t.yaml
- damo_yolo_l.yaml
weights/ – Trained model weights
Naming convention
- MV → Majority Voting
- ST → STAPLE (expected future update)
- YOLOv11-S/
  - UVH-26-MV-YOLOv11-S.pt
  - (ST version coming soon)
- YOLOv11-X/
  - UVH-26-MV-YOLOv11-X.pt
  - (ST version coming soon)
- RT-DETRv2-S/
  - UVH-26-MV-RT-DETRv2-S.pth
  - (ST version coming soon)
- RT-DETRv2-X/
  - UVH-26-MV-RT-DETRv2-X.pth
  - (ST version coming soon)
- DAMO-YOLO-T/
  - UVH-26-MV-DAMO-YOLO-T.pth
  - (ST version coming soon)
- DAMO-YOLO-L/
  - UVH-26-MV-DAMO-YOLO-L.pth
  - (ST version coming soon)

Note — Future Releases

STAPLE (ST) variants for all models will be released soon in a follow-up version.

Model Families

Model Family	Variants	Consensus Versions	Format
YOLOv11	S, X	ST, MV	`.pt`
RT-DETRv2	S, X	ST, MV	`.pth`
DAMO-YOLO	T, L	ST, MV	`.pth`

MV (Majority Voting): Uses strict label agreement for conservative labeling.
ST (STAPLE): Uses probabilistic label fusion for reduced annotation noise.

Classes

The file uvh_classes.txt lists all 14 object categories, one per line:

ID	Class Name	Description
1	Hatchback	Small passenger cars without a protruding rear boot (“dickey”).
2	Sedan	Passenger cars with a low-slung design and a separate protruding rear boot (“dickey”).
3	SUV	Car-like vehicles with high ground clearance, a sturdy body, and no protruding boot.
4	MUV	Large vehicles with three seating rows, combining passenger and cargo functionality.
5	Bus	Large passenger vehicles used for public or private transport, including office shuttles and intercity buses.
6	Truck	Heavy goods carriers with a front cabin and a rear cargo compartment.
7	Three-wheeler	Compact vehicles with one front wheel and two rear wheels, featuring a covered passenger cabin.
8	Two-wheeler	Motorbikes and scooters for single or double riders. Bounding boxes include both vehicle and rider.
9	LCV	Lightweight goods carriers used for short- to medium-distance transport.
10	Mini-bus	Shorter, compact buses with fewer seats; larger than a Tempo Traveller, often featuring a flat front.
11	Tempo-traveller	Medium-sized passenger vans with tall roofs and side windows; larger than vans but smaller than minibuses, with a protruding front.
12	Bicycle	Non-motorized, manually pedalled vehicles including geared, non-geared, women’s, and children’s cycles. Bounding boxes include both vehicle and rider.
13	Van	Medium-sized vehicles for transporting goods or people, typically with a flat front and sliding side doors; smaller than Tempo Travellers.

Training Hyperparameters and Architecture

Setting	DAMO-YOLO-S	DAMO-YOLO-X	YOLOv11-S	YOLOv11-X	RT-DETRv2-S	RT-DETRv2-X
Batch Size	16	16	16	16	16	16
Best Epoch	47	18	30	13	66	68
Learning Rate / Optimizer	0.01 / SGD	0.01 / SGD	auto / AdamW	auto / AdamW	1e-4 / AdamW	1e-4 / AdamW
Weight Decay / Momentum	5e-4 / 0.9	5e-4 / 0.9	5e-4 / (0.9, 0.999)	5e-4 / (0.9, 0.999)	1e-4 / (0.9, 0.999)	1e-4 / (0.9, 0.999)
LR Policy	Constant	Constant	Cosine	Cosine	Linear	Linear
Warm-up / No-Aug Phase	5 epochs / 16 epochs	5 / 16	patience = 150	patience = 150	– / 70 epochs	– / 70 epochs
Augmentation	mixup 0.15, shear 2°, rotate 10°	mixup 0.15, shear 2°, rotate 10°	no mosaic/mixup	no mosaic/mixup	flip, color-distort, zoom, IoU-crop (to 70 epochs)	flip, color-distort, zoom, IoU-crop (to 70 epochs)
Backbone	TinyNAS-L20 + GiraffeNeck V2	TinyNAS-L45 + GiraffeNeck V2	YOLOv11-S + CSP	YOLOv11-X + CSP	PResNet-18 + HybridEnc	PResNet-101 + HybridEnc

All models were trained on the UVH-26 dataset with identical batch sizes and consistent augmentation settings for fair comparison.

License

This repository (models, weights, configs) is released under the Apache License 2.0.
Note: The underlying YOLO-family models (e.g., YOLOv11) from Ultralytics are distributed under the GNU AGPL v3.0 (or newer) license.

SaurabhKrGupta7788/UVH-26