Skip to main content

What are false positives/negatives?

C
Written by Christopher Ren
Updated over a year ago

Reality Defender’s platform contains the most accurate and comprehensive detection models and methods in existence, dwarfing any other solution (especially those focused on using a single detection model or method in their search).

No detection method will ever be 100%. There will occasionally be false positives and false negatives. This means fake videos that are labeled as real or real videos labeled as fake.

The number of false positives will depend on the test scenario. For example, in audio, our models have been tested to work best for studio quality English language samples with American and British accents. When fed with out-of-distribution data, we achieve decent generalization but there could be an increased number of false positives/false negatives. Possible factors include:

a. Multiple speakers in the audio

b. High background noise

c. Accents outside American and British

d. Original compression format of the audio

Note that we are always aggressively working to improve our models.

Did this answer your question?