Open zeynepsu opened 9 years ago
I confirmed that it is the ycbcr classifier that was added to this port of nude.js that causes this discrepancy between nude.js and nude.py. I will be doing more tests to see which one is beneficial to my specific use case.
Hi! I recently discovered this project, looking for software that detects nudity, to analyze a large number of photos of user profiles. I do not program in Python (yet).
I wanted to report the same problem thas was reported by zeynepsu, and even based on her comment I tried disabling the «ycbcr_classifier» in th line nude.py:312, and got better results on a set of 40 images of example, where 20 have nudity and other 20 without it.
In short, to analyze the images and enabling / disabling that classifier, I got the following misclassified images:
image type | Enabled | Disabled |
---|---|---|
nude | 11 | 6 |
no nude | 8 | 5 |
Note: The test images were taken from this very small script witen in Python too: pornscanner
I cloned the project and ran the tests immediately and got "True" for damita2.jpg from the examples. I will investigate to see what causes this.