ONNX NeuFlowV2 Optical Flow Estimation

Python scripts performing optical flow estimation using the NeuFlowV2 model in ONNX.

Requirements

• Check the requirements.txt file.
• For ONNX, if you have a NVIDIA GPU, then install the onnxruntime-gpu, otherwise use the onnxruntime library.

Installation

git clone https://github.com/ibaiGorordo/ONNX-NeuFlowV2-Optical-Flow.git
cd ONNX-NeuFlowV2-Optical-Flow
pip install -r requirements.txt

ONNX model

• If the model file is not found in the models directory, it will be downloaded automatically from the Release Assets.
• Available models: neuflow_mixed.onnx, neuflow_sintel.onnx, neuflow_things.onnx
• You can export a custom model using the Pytorch Inference repository below.

Pytorch Inference / ONNX Export

• Pytorch Inference repository: https://github.com/ibaiGorordo/NeuFlow_v2-Pytorch-Inference
• Includes the minimal code to perform inference with the NeuFlowV2 model in Pytorch.
• You can export your custom model to ONNX using the export_onnx.py script.

Original NeuflowV2 model

The original NeuflowV2 model can be found in this repository: https://github.com/neufieldrobotics/NeuFlow_v2

• The License of the models is Apache-2.0 license: https://github.com/neufieldrobotics/NeuFlow_v2/blob/master/LICENSE

Examples

• Image inference:

python image_optical_flow.py

• Webcam inference:

python webcam_optical_flow.py

• Video inference:

python video_optical_flow.py

• Optical Flow of Detected Objects: https://youtu.be/S0RnlEHGNrc

Install first YOLOv10 detection library: pip install yolov10-onnx

python video_detection_optical_flow.py

References:

• NeuflowV2 model: https://github.com/neufieldrobotics/NeuFlow_v2
• NeuflowV2 Pytorch Inference: https://github.com/ibaiGorordo/NeuFlow_v2-Pytorch-Inference