Toybrick EN

中文 / EN
Toybrick EN Wiki TB-96AI Model Inference
Model Inference

This chapter focuses on how to call the RKNN Python API for model inference on the TB-96AI RK3399Pro development board.

API call flow 



  1. import numpy as np
    from PIL import Image
    from rknn.api import RKNN
    # Analyze the output of the model to get the most probable gesture and corresponding probability
    def get_predict(probability):
        data = probability[0][0]
        data = data.tolist()
        max_prob = max(data)
    return data.index(max_prob), max_prob;
    def load_model():
        # Create an RKNN execution object
        rknn = RKNN()
        # Load RKNN model
        print('-->loading model')
        print('loading model done')
        # Initialize the RKNN runtime environment
        print('--> Init runtime environment')
        ret = rknn.init_runtime(host='rk3399pro')
        if ret != 0:
           print('Init runtime environment failed')
        return rknn
    def predict(rknn):
        im ="../picture/6_7.jpg")   # load image
        im = im.resize((64, 64),Image.ANTIALIAS)  # Image resize to 64x64
        mat = np.asarray(im.convert('RGB'))    # Convert to RGB format
        outputs = rknn.inference(inputs=[mat])   # Run forward inference and get the inference result
        pred, prob = get_predict(outputs)     # Transform the inference results into visual information
    if __name__=="__main__":
        rknn = load_model()

Products Store Community Wiki Download About TB

To Top