Posted in Uncategorized

Feature-request: State-of-art Yolo v4 Detector

   September 23, 2020  1 Comment on Feature-request: State-of-art Yolo v4 Detector

Feature-request: State-of-art Yolo v4 Detector

Many other features from Darknet were added previously.
But should be added at least new two features:

  1. Mish-activation:
float softplus(float x, float threshold = 20) {
    if (x > threshold) return x;                // too large
    else if (x < -threshold) return expf(x);    // too small
    return logf(expf(x) + 1);
}

float mish_activation(float input) {
    const float MISH_THRESHOLD = 20;
    output = input * tanh( softplus(input, MISH_THRESHOLD) );
    return output;
}
  1. Eliminate grid sensitivity – by using scale_x_y= parameter from [yolo] layer in cfg-file (by default use scale_x_y=1.0 if the value is not set in cfg file)
    we should use:
const float x_tmp = logistic_activate(srcData[box_index + 0]) * scale_x_y - (scale_x_y - 1) / 2;
const float y_tmp = logistic_activate(srcData[box_index + 1]) * scale_x_y - (scale_x_y - 1) / 2;
dstData[box_index + 0] = (x + x_tmp)) / cols;
dstData[box_index + 1] = (y + y_tmp)) / rows;

instead of

opencv/modules/dnn/src/layers/region_layer.cpp

Lines 305 to 306
in
51a42c0

dstData[box_index + 0] = (x + logistic_activate(srcData[box_index + 0])) / cols;
dstData[box_index + 1] = (y + logistic_activate(srcData[box_index + 1])) / rows;

Comparison of detectors on Microsoft COCO dataset and Tesla V100 GPU:

comparison

modern_gpus

Full dicussion on github.com

1 possible answer(s) on “Feature-request: State-of-art Yolo v4 Detector”