Synthetic✦ Production-ReadyAutomotiveDetectionADAS

ADAS Detection - v1

Photorealistic synthetic dataset simulating multi-FOV fisheye camera setups on trucks (front, side, rear), capturing real-world fleed mounting conditions. Designed for object detection in high-variance trucking environments where traditional car-based ADAS datasets fail. Enables robust model training and evaluation across unique perspectives, distortions, and edge-case scenarios specific to fleet operations.

Request Custom Synthetic Dataset

Sample Frames

20 annotated samples drawn from the train split. Toggle annotations to inspect bounding-box quality.

All images in this dataset are 100% synthetically generated. No real-world footage was used.

Class Distribution

14 annotation classes · 773 total images. Sorted by object count, descending.

Annotation counts

Class Balance

Per-class counts, frequency, and average bounding-box area. Sort any column to surface the rarest or most prevalent classes for re-balancing.

Class	Images↓ with class	Objects total	Per image average	Area % of frame
person	746	3,339	4.48	0.56%
car	643	7,540	11.73	0.41%
bicycle	542	1,002	1.85	0.88%
truck	386	1,468	3.80	1.45%
fire_hydrant	224	236	1.05	0.31%
bench	218	238	1.09	1.30%
motorcycle	141	146	1.04	1.11%
bus	123	134	1.09	3.31%
traffic_light	115	468	4.07	0.10%
stop_sign	113	121	1.07	0.36%
parking_meter	109	128	1.17	0.36%
train	24	29	1.21	5.11%
boat	18	25	1.39	1.72%
airplane	17	17	1	0.07%

Co-occurrence Matrix

How frequently pairs of classes appear in the same image. Diagonal cells show standalone image count for that class. Useful for spotting biased or correlated labels.

Pair frequency

	person	bicycle	car	motorcycle	airplane	bus	train	truck	boat	traffic_light	fire_hydrant	stop_sign	parking_meter	bench
person	746	537	620	141	15	118	22	370	18	111	218	106	106	211
bicycle	537	542	459	98	13	83	16	256	15	90	188	63	79	175
car	620	459	643	127	15	114	10	296	17	103	194	103	101	189
motorcycle	141	98	127	141	5	20	2	71	6	15	28	22	16	27
airplane	15	13	15	5	17	2	1	8	1		1	1	1	2
bus	118	83	114	20	2	123	5	43	1	33	40	20	12	14
train	22	16	10	2	1	5	24	15	1	1		2	2	5
truck	370	256	296	71	8	43	15	386	14	50	86	43	36	83
boat	18	15	17	6	1	1	1	14	18	3	2	3	3	3
traffic_light	111	90	103	15		33	1	50	3	115	38	10	14	33
fire_hydrant	218	188	194	28	1	40		86	2	38	224	24	32	62
stop_sign	106	63	103	22	1	20	2	43	3	10	24	113	14	23
parking_meter	106	79	101	16	1	12	2	36	3	14	32	14	109	27
bench	211	175	189	27	2	14	5	83	3	33	62	23	27	218

Hover any cell for image count0620

Average Object Area

Each rectangle is one class, sized by the average area its bounding boxes occupy as a percentage of the frame. Surfaces tiny vs. dominant objects at a glance.

% of frame

Spatial Distribution

Where annotations of each class tend to fall across the frame. Brighter regions indicate higher density — useful for detecting positional bias in your training data.

Per-class heatmaps

person3,339

bicycle1,002

car7,540

motorcycle146

airplane17

bus134

train29

truck1,468

boat25

traffic_light468

fire_hydrant236

stop_sign121

parking_meter128

bench238

lowhigh density

Model Performance

Validation metrics from a YOLOv8 detector trained on this dataset. Reference checkpoint: yolov8m.pt (best.pt).

Validation set

These results are based on training on 100% synthetic images from this dataset, validated on 100% real-world held-out images. For production deployments, AnywayLabs.ai recommends mixing 10–25% real images into your training set.

0.7025

mAP@0.5

0.4796

mAP@0.5:0.95

0.9620

Precision

0.6508

Recall

Per-class AP

Class	AP@0.5	AP@0.5:0.95	Precision	Recall
person	0.7685	0.4645	0.9850	0.6349
bicycle	0.6689	0.5318	0.9818	0.6667

Validation curves over training

mAP@0.5

mAP@0.5:0.95

Precision

Recall

Dataset Metadata

Annotation format	YOLO
Total images	773
Classes	14
Resolution	1920x1080
Last updated	2026-05

Interested in this dataset?

Get in touch for sample access, custom variants, or licensing terms — typical turnaround is two business days.

Request Custom Synthetic Dataset