Skip to content

Latest commit

 

History

History
44 lines (28 loc) · 1.35 KB

README.md

File metadata and controls

44 lines (28 loc) · 1.35 KB
Raw

Training Object detection model + Epic Kitchens 55 Dataset

Dataset information

You can get the dataset information in EPIC-KITCHENS-55.

Download dataset

Build docker image

Then, you need to build docker image called train-epic-kitchens-55-edgetpu-object-detection.

roscd coral_usb/training/epic_kitchens_55
make

Train model

Finally, you can train the model with your dataset.

roscd coral_usb/training/epic_kitchens_55
bash ./run.sh <your_dataset_path> <annotation_path>

The dataset can be downloaded with epic-kitchens/epic-kitchens-download-scripts. Annotation can be found in epic-kitchens/epic-kitchens-55-annotations.

After training, you can get trained model and label file as below:

  • EdgeTPU model: <your_dataset_path>/learn/models/output_tflite_graph_edgetpu.tflite
  • Label file: <your_dataset_path>/learn/models/labels.txt

Visualize your training result with TensorBoard

You can visualize your training result with TensorBoard.

TensorBoard port is often set around 6006.

roscd coral_usb/training/epic_kitchens
bash ./run.sh <your_dataset_path> --port <port> tensorboard