[?] Is a threshold applied to predictions in evaluation?

[rizavelioglu]

The model makes 100 predictions per sample. For some settings, e.g. a single object in an image, 100 predictions could yield a lot of wrong predictions, possibly with low confidence.
Therefore, is a threshold applied to the model's output, for example, removing any prediction with a confidence score of less than 0.3 (like the argument score-thr used in demo code)?
More specifically, are the provided results in Tables 2 and 3 based on the thresholded output or the raw output (100 predictions)?