| Model | KITTI 2012 D1 (%) | KITTI 2015 D1 (%) | ETH3D Bad1.0 (%) | Middlebury (H) Bad2.0 (%) | DrivingStereo Weather D1 (%) |
|---|---|---|---|---|---|
| LAS2-M | 2.88 | 3.61 | 2.59 | 5.47 | 7.67 |
| LAS2-L | 2.57 | 3.38 | 1.83 | 5.28 | 7.98 |
[[1, 1, 800, 1280], [1, 1, 800, 1280]] • Output shape: [1, 1, 800, 1280]27.029 • GFLOPs: 392.230| Platform | Precision | Throughput (infs/sec) | Power Consumption (W) |
|---|---|---|---|
| RVC4 | FP16 | 6.26 | 3.43 |
[[1, 1, 800, 1280], [1, 1, 800, 1280]] • Output shape: [1, 1, 800, 1280]27.033 • GFLOPs: 392.376| Platform | Precision | Throughput (infs/sec) | Power Consumption (W) |
|---|---|---|---|
| RVC4 | MP | 10.48 | 3.49 |
[[1, 1, 640, 1024], [1, 1, 640, 1024]] • Output shape: [1, 1, 640, 1024]27.035 • GFLOPs: 251.240| Platform | Precision | Throughput (infs/sec) | Power Consumption (W) |
|---|---|---|---|
| RVC4 | MP | 15.47 | 3.99 |
[[1, 1, 480, 768], [1, 1, 480, 768]] • Output shape: [1, 1, 480, 768]10.366 • GFLOPs: 63.822| Platform | Precision | Throughput (infs/sec) | Power Consumption (W) |
|---|---|---|---|
| RVC4 | MP | 41.52 | 3.68 |
[[1, 1, 320, 512], [1, 1, 320, 512]] • Output shape: [1, 1, 320, 512]10.369 • GFLOPs: 28.393| Platform | Precision | Throughput (infs/sec) | Power Consumption (W) |
|---|---|---|---|
| RVC4 | MP | 94.14 | 3.56 |
pip install --upgrade depthai depthai-nodes
import depthai as dai
from depthai_nodes.node import ParsingNeuralNetwork
model_description = dai.NNModelDescription(
"luxonis/lite-any-stereo-v2:l-fp16-1280x800", platform="RVC4"
)
width, height = 1280, 800
with dai.Pipeline() as pipeline:
left = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_B, sensorFps=30)
right = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_C, sensorFps=30)
stereo = pipeline.create(dai.node.StereoDepth).build(
left.requestOutput((width, height), type=dai.ImgFrame.Type.GRAY8, fps=2),
right.requestOutput((width, height), type=dai.ImgFrame.Type.GRAY8, fps=2),
)
stereo.setRectifyEdgeFillColor(0)
nn_with_parser = pipeline.create(ParsingNeuralNetwork)
nn_with_parser.setNNArchive(dai.NNArchive(dai.getModelFromZoo(model_description)))
# Link rectified frames to neural network
for name, output in [("left", stereo.rectifiedLeft), ("right", stereo.rectifiedRight)]:
nn_with_parser.inputs[name].setBlocking(True)
nn_with_parser.inputs[name].setMaxSize(4)
output.link(nn_with_parser.inputs[name])
parser_output_queue = nn_with_parser.out.createOutputQueue()
pipeline.start()
while pipeline.isRunning():
parser_output: dai.beta.Map2D = parser_output_queue.get()
disparity = parser_output.getMap()
Ultra-fast stereo matching model designed for efficient zero-shot stereo matching. | |
License | MIT Commercial use |
Downloads | 1614 |
Tasks | Depth Estimation |
Model Types | ONNX |
| Name | Version | Available For | Created At | Deploy |
|---|---|---|---|---|
| RVC4 | About 2 months ago | |||
| RVC4 | About 2 months ago | |||
| RVC4 | About 2 months ago | |||
| RVC4 | About 2 months ago | |||
| RVC4 | About 2 months ago |