X

Model Details

Model Description

YOLO-P is a panoptic driving perception network that performs traffic object detection, drivable area segmentation, and lane detection simultaneously. It is composed of one encoder for feature extraction and three decoders to handle the specific tasks.

Developed by: Dong Wu et al.
Shared by:
Model type: Computer Vision
License:
Resources for more information:

Training Details

Training Data

Trained and evaluated on the dataset. The BDD100K dataset has three parts, a training set with 70K images, a validation set with 10K images, and a test set with 20K images. The evaluation of the model was conducted in the validation set, since the label of the test set is not public.

Testing Details

Metrics

These results showcase the performance of a multi-task model on the BDD100K dataset across three key autonomous driving tasks: Traffic Object Detection, Drivable Area Segmentation, and Lane Detection. The results are obtained from the original .

Traffic Object Detection Results

Dataset	Recall (%)	mAP50 (%)
BDD100K	89.2 (+1.0)	76.5 (-0.4)

Drivable Area Segmentation Results

Model	mIOU (%)
BDD100K	91.5 (-0.1)

Lane Detection Results

Dataset	Accuracy (%)	IOU (%)
BDD100K	70.50 (+0.6)	26.20 (-0.3)

The values in parentheses represent confidence intervals or uncertainty measures, indicating the statistical reliability of each reported metric. These intervals provide a range within which the true value is likely to fall, accounting for variability in the measurements.

Technical Specifications

Input/Output Details

Input:
- Name: images
  - Info: NCHW, BGR un-normalized image
Output:
- Name: output1_yolop
  - Info: NCHW, first detection head
- Name: output2_yolop
  - Info: NCHW, second detection head
- Name: output3_yolop
  - Info: NCHW, third detection head
- Name: drive_area_seg
  - Info: NCHW, output of the drivable area segmentation head
- Name: lane_line_seg
  - Info: NCHW, output of the lane line segmentation head

Model Architecture

YOLOP is a single-shot network which contains one shared encoder and three subsequent decoders to solve specific tasks. It comprises of the Backbone witch is the encoder, the Neck which is used to fuse the features generated by the backbone, and the three heads, Detect Ηead, Drivable Area Segment Ηead, and Lane Line Segment Ηead. See the for more information.

Throughput

Model variant: yolo-p:bdd100k-320x320

• Input shape: [1, 3, 320, 320] • Output shapes: [[1, 18, 40, 40], [1, 18, 20, 20], [1, 18, 10, 10], [1, 2, 320, 320], [1, 2, 320, 320]]

• Params (M): 7.936 • GFLOPs: 4.051

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	15.61	N/A
RVC4	INT8	374.06	4.37

* Benchmarked with , using 2 threads (and the DSP runtime in balanced mode for RVC4).

* Parameters and FLOPs are obtained from the package.

Utilization

Models converted for RVC Platforms can be used for inference on OAK devices. DepthAI pipelines are used to define the information flow linking the device, inference model, and the output parser (as defined in model head(s)). Below, we present the most crucial utilization steps for the particular model. Please consult the docs for more information.

Install DAIv3 and depthai-nodes libraries:

pip install depthai
pip install depthai-nodes

Define model:

model_description = dai.NNModelDescription(
    "luxonis/yolo-p:bdd100k-320x320"
)

nn = pipeline.create(ParsingNeuralNetwork).build(
    <CameraNode>, model_description
)

Inspect model head(s):

YOLOExtendedParser that outputs message (detected cars).
SegmentationParser that outputs message (segmentation mask of drivable area).
SegmentationParser that outputs message (segmentation mask of lane lines).

The model is multi-headed. You can set up the queues as follows:

detection_parser_output_queue = nn_with_parser.getOutput(0).createOutputQueue()
da_seg_parser_output_queue = nn_with_parser.getOutput(1).createOutputQueue()
ll_seg_parser_output_queue = nn_with_parser.getOutput(2).createOutputQueue()

Get parsed output(s):

while pipeline.isRuning():
    detection_parser_output: ImgDetectionsExtended = detection_parser_output_queue.get()
    da_seg_parser_output: SegmentationMask = da_seg_parser_output_queue.get()
    ll_seg_parser_output: SegmentationMask = ll_seg_parser_output_queue.get()

Example

You can quickly run the model using our example.

The example demonstrates how to build a 1-stage DepthAI pipeline consisting a road segmentation model. It automatically downloads the model, creates a DepthAI pipeline, runs the inference, and displays the results using our DepthAI visualizer tool.

To try it out, run:

python3 main.py

You Only Look at Once for Panoptic driving Perception
License	MIT Commercial use
Downloads	1005
Tasks	Semantic Segmentation
Model Types	ONNX