4 ms·

CSAM is the author's baseless speculation, but I could find no evidence suggesting NudeNet contains any. That it's hosted on GitHub, all its files are on archiv

CSAM is the author's baseless speculation, but I could find no evidence suggesting NudeNet contains any. That it's hosted on GitHub, all its files are on archive.org, and is a widely used data-set 6 years old by now, all indicate that it doesn't contain CSAM. In general, when a company does something hostile and unreasonable, let's not invent out of thin air facts to excuse their behavior, shall we? Especially since their stonewalling is deliberate.

markatlarge 1y ago
Google won’t tell me the name of the file—or files—they believe violated their policy. So IT IS baseless speculation. I only know what I was doing prior to the account deletion—and I’m guessing that’s what triggered the issue. Google should at minimum publish the filenames of the content they consider violations. I have Colab notebooks for everything I’ve downloaded from the internet. With a filename, I could trace the origin and have the image independently reviewed. I suspect it wasn’t CSAM at all. We know nothing about how their detection system or hashing works. I’d also like to point out that I benchmarked my model against Google’s own commercially available NSFW detector—and mine actually performed better: https://medium.com/@russoatlarge/benchmark-study-punge-yolov11n-on-device-vs-google-cloud-vision-for-nsfw-image-detection-d4462217ed15 https://medium.com/@russoatlarge/benchmark-study-punge-yolov... As for datasets: COCO is the most widely used dataset in computer vision, with hundreds of thousands of labeled images. One image in the COCO Train2017 set—000000001790.jpg—shows a child brushing his teeth while looking in the mirror in his “birthday suit.” But if Google's system flagged that as CSAM, then every researcher using COCO is at risk. If Google genuinely cared about addressing CSAM responsibly, they would share the filenames so they could be independently verified and—if needed—removed from circulation. Instead, they’re silent. That silence creates fear, not safety.

markatlarge 1y ago

Google won’t tell me the name of the file—or files—they believe violated their policy. So IT IS baseless speculation. I only know what I was doing prior to the account deletion—and I’m guessing that’s what triggered the issue. Google should at minimum publish the filenames of the content they consider violations. I have Colab notebooks for everything I’ve downloaded from the internet. With a filename, I could trace the origin and have the image independently reviewed. I suspect it wasn’t CSAM at all. We know nothing about how their detection system or hashing works. I’d also like to point out that I benchmarked my model against Google’s own commercially available NSFW detector—and mine actually performed better: https://medium.com/@russoatlarge/benchmark-study-punge-yolov11n-on-device-vs-google-cloud-vision-for-nsfw-image-detection-d4462217ed15 https://medium.com/@russoatlarge/benchmark-study-punge-yolov... As for datasets: COCO is the most widely used dataset in computer vision, with hundreds of thousands of labeled images. One image in the COCO Train2017 set—000000001790.jpg—shows a child brushing his teeth while looking in the mirror in his “birthday suit.” But if Google's system flagged that as CSAM, then every researcher using COCO is at risk. If Google genuinely cared about addressing CSAM responsibly, they would share the filenames so they could be independently verified and—if needed—removed from circulation. Instead, they’re silent. That silence creates fear, not safety.