257 - Exploring GAN latent space to generate images with desired features​

Code generated in the video can be downloaded from here: https://github.com/bnsreenu/python_fo... UTKFace dataset (Used in this video): https://susanqq.github.io/UTKFace/ Haarcascade models, if interested in detecting faces and extracting them into new images. https://github.com/opencv/opencv/tree... Celeb Dataset (Not used in the video): https://www.kaggle.com/jessicali9530/... Description: Latent space is hard to interpret unless conditioned using many classes.​ But, the latent space can be exploited using generated images.​ Here is how... Generate 10s of images using random latent vectors.​ Identify many images within each category of interest (e.g., smiling man, neutral man, etc. )​ Average the latent vectors for each category to get the mean representation in the latent space (for that category).​ Use these mean latent vectors to generate images with features of interest. ​ In summary, you can find the latent vectors for Smiling Man, neutral face man, and a baby with a neutral face and then generate a smiling babyface by: Smiling Man + Neutral Man - Neutral baby = Smiling Baby