Should we apply scale/offset to statistics?

[vincentsarago]

rio-tiler/rio_tiler/reader.py

Lines 271 to 277 in c9aaaf0

	stats = []
	for ix in indexes:
	tags = dataset.tags(ix)
	if all(stat in tags for stat in ["STATISTICS_MINIMUM", "STATISTICS_MAXIMUM"]):
	stat_min = float(tags.get("STATISTICS_MINIMUM"))
	stat_max = float(tags.get("STATISTICS_MAXIMUM"))
	stats.append((stat_min, stat_max))

We allow users to apply scale/offset when reading the data (with unscale=True option), if present in the metadata, the stats will be forwarded to the ImageData object, if scale/offset was applied then the statistics won't make sense.

This is True if we assume the stats are computed with the raw pixel values, not with scale/offset applied.