As many people have suggested, it would be interesting for the community to create an image dataset to improve the algorithm.
To achieve success, we must begin by asking the right questions:
1- Are we able to train deep learning ourselves if we have a better dataset?
2- What type of image (quality, quantity, resolution, format, content) would it be necessary to improve the current algorithm?
==> About the content: What kind of photo should we provide? dressed and naked? or just naked?
3- Can we reasonably launch such a project without web support (centralized website)?
Of course, other questions will arise from this first reasoning. The idea here is to think about whether such a project could be carried out.
Answers are welcome!