RF-DETR Nano Instance Segmentation

Our new model ZOO works with DepthAI V3. Find out more in our documentation.

Model Details

Model Description

RF-DETR Nano Instance Segmentation is a real-time transformer-based instance segmentation model from Roboflow's RF-DETR family. It is the smallest segmentation variant in the lineup, designed for low-latency inference while retaining strong COCO accuracy. The model is built on a DINOv2-based vision transformer backbone and is exported here at 312x312 input resolution.

Developed by: Roboflow

Shared by:

Model type: Computer Vision

License:

Resources for more information:

To deploy RF-DETR to Luxonis RVC4 devices some modifications need to be made to the model architecture. This are mainly there to make the operations compatible with the device accelerator. Please refer to to learn more about this.

Training Details

Training Data

The base RF-DETR Nano Instance Segmentation checkpoint is trained on the dataset for instance segmentation. This export uses the COCO label space and predicts both bounding boxes and per-instance masks.

The exported parser metadata contains 91 COCO category slots (including background and unused category IDs), which matches the standard COCO indexing used by the packaged parser, while the underlying task covers the standard 80 foreground COCO classes.

Testing Details

Metrics

The following benchmark numbers are taken from the . COCO metrics are reported for the validation split.

Model	COCO AP50	COCO AP50:95	Latency (ms)	Params ^(M)
RF-DETR Nano Instance Segmentation	63.0	40.3	3.4	33.6

Technical Specifications

Input/Output Details

Input:

Name: input

Info: NCHW RGB image tensor of shape [1, 3, 312, 312]

Outputs:

Name: dets

Info: Bounding box tensor of shape [1, 100, 4] for up to 100 detections

Name: labels

Info: Per-detection COCO class score tensor of shape [1, 100, 91]

Name: masks

Info: Per-instance mask tensor of shape [1, 100, 78, 78]

Model Architecture

Backbone: DINOv2 vision transformer encoder

Task: Joint object detection and instance segmentation

Head: RF-DETR segmentation head with up to 100 object queries exposed in the exported parser configuration

Throughput

Model variant: rf-detr-nano-instance-segmentation:coco-384x384

• Input shape: [1, 3, 312, 312] • Output shapes: [[1, 100, 4], [1, 100, 91], [1, 100, 78, 78]]

• Params (M): 30.597 • GFLOPs: 24.443

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC4	FP16	27.70	3.41

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

* Parameters and FLOPs are obtained from the exported ONNX files with the package after running ONNX shape inference on the exported graph.

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. Below, we present the most crucial utilization steps for this model. Please consult the docs for more information.

Install DAIv3 library:

pip install depthai
pip install depthai-nodes

Define model:

model_description = dai.NNModelDescription(
    "luxonis/rfdetr-nano-instance-segmentation:coco-312x312"
)

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

Inspect model head(s):

RFDETRParser that outputs message containing bounding boxes, labels, confidence scores for the detected objects and their instance segmentation mask.

Get model output(s):

while pipeline.isRunning():
    output = parser_output_queue.get()

Example

You can quickly run the model using the script. It automatically downloads the model, creates a DepthAI pipeline, runs the inference, and displays the results using the DepthAI visualizer tool. To try it out, run:

python3 main.py \
    --model luxonis/rfdetr-nano-instance-segmentation:coco-312x312

RF-DETR is a real-time transformer architecture for object detection and instance segmentation developed by Roboflow.
License	Apache 2.0 Commercial use
Downloads	18
Tasks	Instance Segmentation
Model Types	ONNX

Model Variants

Name	Version	Available For	Created At	Deploy
		RVC4	12 days ago