Hi @L-YeZhu
I use the boundary search on SD-v1.4 model, with latent of shape (4,64,64). I sampled in total 200 pairs of images. The training for SVM is extremely slow, over 20mins
But I noticed that in the paper, "In practice, the hyperplanes are found via linear SVMs [17 ], with almost negligible learning time of about 1 second on a single RTX3090 GPU." I also use the svm in sklearn, which does not support GPU. I am wondering if there is something wrong with the configuration?
Hi @L-YeZhu
I use the boundary search on SD-v1.4 model, with latent of shape (4,64,64). I sampled in total 200 pairs of images. The training for SVM is extremely slow, over 20mins
But I noticed that in the paper, "In practice, the hyperplanes are found via linear SVMs [17 ], with almost negligible learning time of about 1 second on a single RTX3090 GPU." I also use the svm in sklearn, which does not support GPU. I am wondering if there is something wrong with the configuration?