Training of Object Detection model

This Python script allows you to train a Pytorch Object Detection model with preprocessed HDF5 data and labeled bounding boxes.

The source code of this sample can be found in <install-prefix>/share/metavision/sdk/ml/python_samples/train_detection when installing Metavision SDK from installer or packages. For other deployment methods, check the page Path of Samples.

Expected Output

After completing the training process, the following outputs are generated:

  • Checkpoints: saved models at various stages of training, allowing you to resume training, fine-tune, or evaluate the model at specific points in its learning process.

  • Log Files: detailed logs which can be used to monitor training and troubleshoot issues.

  • Validation Videos: visual outputs showcasing the model’s performance on the test dataset

A test demo is shown below:



Setup & requirements

To run the script, you need:

  • path to the output folder

  • path to the training dataset:

    • a folder containing 3 sub folders, named train, val, test.

    • each subfolder should contain one or multiple h5 files and their corresponding <h5 filename>_bbox.npy labels.

    • a dictionary file named label_map_dictionary.json, which contains all the detection categories.

Note

To have a quick test of the pipeline, you can simply set the path of training dataset to “toy_problem”. This will train the model on Moving MNIST Dataset.

How to start

To run the script:

python train_detection.py /path/to/logging /path/to/dataset

To find the full list of options, run:

python train_detection.py -h