Pre-trained Models

We provide pre-trained models for object detection and optical flow estimation.

These models can be used to run our ML samples.

The object detection model was trained on a dataset recorded by a camera positioned on top of a car facing forward, thus the performance of the model can be degraded in other settings.

Pre-trained Models in Essentials Offer

For Essentials users, pre-trained models can be downloaded during the sign-up.

Those models are object detection PyTorch Jit models (the only models that can be used with C++ metavision_detection_and_tracking_pipeline)

  • red_histogram_05_2020: Object Detection PyTorch Jit model, trained with histogram preprocessing method

  • red_event_cube_05_2020: Object Detection PyTorch Jit model, trained with event_cube preprocessing method

Pre-trained Models in the Professional Offer

For Professional users, pre-trained models can be found in the source code of ML module in

  • metavision_professional_ml/sdk/modules/ml/python/models -> object detection PyTorch Jit models (the only models that can be used with C++ metavision_detection_and_tracking_pipeline)

    • red_histogram_05_2020: Object Detection PyTorch Jit model, trained with histogram preprocessing method

    • red_event_cube_05_2020: Object Detection PyTorch Jit model, trained with event_cube preprocessing method

  • metavision_professional_ml/sdk/modules/ml/python_extended/models