Pre-trained Models
We provide some models for Core ML and ML modules.
If you are compiling OpenEB, the models for the Core ML module can be found in
<SRC_DIR>/sdk/modules/core_ml/models.If you are compiling SDK from sources, the Core ML models are available at the path above, and some extra models for ML module can be found in
<SRC_DIR>/sdk/modules/ml/models.If you installed Metavision SDK with the packages/installer, you can download the dedicated stand-alone archive containing models of the CORE ML and ML modules. The link to this models archive is listed in the page for your SDK version in the Knowledge Center Download section.
See also
To evaluate these models, you can follow the links to the corresponding code samples. For a comprehensive explanation on how to integrate these models within the SDK, please refer to our :sing ML models page.
Event-to-Video Model
e2v.ckpt: Pytorch Event to Video pre-trained model which can be used in event to video inference sample The model was trained on a dataset of synthetic events sequences generated by simulation on randomly moving images.
Corner Detection Model
corner_detection_10_heatmaps.ckpt: Pytorch Corner Detection pre-trained model which can be used in Demo of Corner Detection Model The model was trained on a dataset of synthetic events sequences generated by simulation on randomly moving images. It takes as input a tensor of 10 channels and outputs 10 heatmaps.
Object Detection Models
red_event_cube_05_2020.zip: Object Detection TorchScript model, trained withevent_cubepreprocessing method. This model can be used with the C++ Detection & Tracking sample, the C++ Detection sample, and the Python Detection & Tracking sample. This 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.
red_event_cube_all_classes.ckpt: Pytorch object detection model, which can be exported to torchjit model using export_detector Python sample. Note that this model was not trained as much as the TorchScript models listed above, and is mostly delivered to try-out the our export sample.
Flow Models
flow_model_alpha.ckpt: Pytorch flow recurrent network which can be used in Flow Inference sample and be exported to torchjit model using export_flow Python sample.
model_flow.ptjit: Torchjit optical flow model obtained from conversion offlow_model_alpha.ckpt. It requires a single event cube as input and can be used in the C++ Optical Flow inference pipeline.
model_flow.onnx: ONNX feed forward optical flow model taking two consecutive histograms as input. It can be used in the C++ Optical Flow inference pipeline.
Classification Models
Pre-Trained models for rock-paper-scissors classification dataset:
convRNN_chifoumi.zip: ConvRNN TorchScript classification which can be used with our Python and C++ Classification Inference samples.
mobilenetv2_chifoumi.zip: Mobilenetv2 TorchScript classification model which can be used with our Python and C++ Classification Inference samples.
mobilenetv2_chifoumi_histo_quantized.zip: Mobilenetv2 TorchScript classification model trained with quantized histograms which can be used with our Python and C++ Classification Inference samples.
chifoumi_rnn.ckpt: ConvRNN Pytorch classification model, which can be exported to torchjit model using export_classifier Python sample.
chifoumi_fnn.ckpt: Mobilenetv2 feed forward Pytorch classification model trained with quantized histograms, which can be exported to ONNX model using export_fnn_classifier_onnx Python sample.