Question about Yolov2-tiny model compilier

  • I have a question about converting darknet yolov2 to kendryte C mode.

    The input of yolov2-tiny is 416x416 in cfg file:


    Meanwhile, the input size of kendryte-model-compiler/pb_files/20classes_yolo.pb is 320x240 (using tensorboard):


    I have two plans to solve the problem. Which one is better? Or is there a better way?

    1. resize/remap the dvp output from 320x240 to 416x416. But it will increase the burden at the runtime.
    2. re-training yolov2-tiny model with 320x240 input.

    Thank you very much for your suggestions!

  • Hello, choose the second way is better.