ML_YOLO

• This tool assists you in training the YOLO-series models using Darknet and converting them to TFLite models, which are easy to deploy on MCU/MPU devices. It also provides support for the Vela compiler for ARM NPU devices.

1. Yolo-Fastest-darknet

• Yolo-Fastest-darknet/
• This is the training model step/workfolder.
• Use Darknet to train the YOLO-Fastestv1 model, which is obtained from dog-qiuqiu/Yolo-Fastest and comes with a pre-trained model.
• How to use it, please check Yolo-Fastest-darknet/.

2. darknet_tflite

• darknet_tflite/
• This step involves converting the Darknet model to a TensorFlow model and then to a TFLite model.
• To handle this step, Python scripts are used, so it is necessary to have our NuEdgeWise environment installed.

3. vela

• vela
• If the user intends to deploy the model on an ARM NPU device, the Vela compiler assists in converting the TFLite model to Vela TFLite C++ source files.

Inference code

• MCU: M55M1
  • ObjectDetection_FreeRTOS/
• MPU: MA35D1