parameter min_bbox_width, minimum_bounding_box_height, minimum_height, minimum_width

I assume the unit of those parameters min_bbox_width, minimum_bounding_box_height, minimum_height, minimum_width, etc. in the training configuration file is pixels.

I wonder, if it is the number of pixels in the original image or in the resized image that is defined in
preprocessing {
output_image_width: 608
output_image_height: 608
min_bbox_width: 1.0
min_bbox_height: 1.0
output_image_channel: 3
}

Thanks!

Hi,
See tlt doc, minimum_bounding_box_height means “Minimum height in pixels to consider as a valid detection post clustering”.