The trick to blacklist the bad samples so that the computation is never ran again for these samples is very neat. I think we should extend it to the safe samples. In many cases, the verification can be long and costly (machine learning analysis of the image, etc.) and should only be done once when the sample is first loaded.
The trick to blacklist the bad samples so that the computation is never ran again for these samples is very neat. I think we should extend it to the safe samples. In many cases, the verification can be long and costly (machine learning analysis of the image, etc.) and should only be done once when the sample is first loaded.