reduce pixel_density problem severity: error -> warning, #339
Conversation
| if v is None or v <= 72: | ||
| self.report.add_error("Image %s: %s (%s pixels per %s) is too low" % (f.ID, k, v, exif.resolutionUnit)) | ||
| self.report.add_warning("Image %s: %s (%s pixels per %s) is too low" % (f.ID, k, v, exif.resolutionUnit)) |
There was a problem hiding this comment.
I don't think this is the right thing to do here. This can be a significant problem for processors, so there should be a loud and clear error.
Moreover, core should at least provide a running definition of suspiciously low DPI – not just in the validator, but in OcrdExif too, and not just in isolation, but in relation to the pixel resolution as well.
There was a problem hiding this comment.
IMHO the best way to react in OcrdExif is to raise an error and set the density to 1 (as if nothing had been set) – all under the assumption of OCR-D/spec#129 that correct settings cannot be enforced in the first place.
|
I'll merge this now for the 2.0.0 release but as with OCR-D/spec#129 the discussion shall continue! @bertsky @wrznr feel free to open an issue here or in spec to discuss the rules of the heuristics and how we should handle bad input images. |
OCR-D/spec#129
OCR-D/assets#33