Closed cwh01 closed 4 weeks ago
Oh yes good point!
I did this for rare cases when the detection's have a size larger than 20% of the original images. As my images were taken from a certain distance I do assume the fruits are small. If you have close up images from a fruit you probably should ignore it or increase the threshold
I wonder why this step is being made. Because I didn't do this step when I was training my own dataset, the results were poor.
![Uploading edc20a03c887a9111f2a52aa2b20ef7c.png…]()