Incorrect NSFW Classification for Civilization VI Leaders' Images

Hi,

I'm experiencing an issue where images of leaders from Civilization VI, like Chandragupta and Alexander the Great, are mistakenly flagged as NSFW by the model. Interestingly, similar images of shirtless men don't trigger the same error. Could you please look into this?

Thanks,

I used those images for test and it's been flagged as NSFW:

those will not trigger:

Strangely, it seems that combining the two images above triggers the NSFW detection.

I've also discovered a certain type of meme that consistently triggers the NSFW detection

No source files or tests are named. Start by reproducing the model's classifications with the linked Civilization VI, shirtless-men, combined, and meme images, then trace the model and training-data entry points to identify the false-positive cause. Done means the reported images no longer receive incorrect NSFW classifications without regressing the comparison cases.

Written by the indexing model from the issue text.