Great work!
I was trying to train the model from scrach on sparse images. The images are binary and most of the pixels are zero. The dataset itself contains roughly 10k samples and the model mar_base was trained for 400 epochs. However, after training, the model still produce noise (colorful patches) though loss is low (~0.05). Could you provide some guesses of how this might be? Have you tried training the model on some other datasets?
Great work!
I was trying to train the model from scrach on sparse images. The images are binary and most of the pixels are zero. The dataset itself contains roughly 10k samples and the model mar_base was trained for 400 epochs. However, after training, the model still produce noise (colorful patches) though loss is low (~0.05). Could you provide some guesses of how this might be? Have you tried training the model on some other datasets?