I am glad to present an implementation of DeepMind’s new “Perceiver: General Perception with Iterative Attention” Model which builds on top of Transformers but solves the quadratic scaling problem without making any assumptions of the data like the previous approaches in TensorFlow. This means you can use the same model on images, audio, videos, etc! This model also achieves state-of-the-art for some tasks!
Find it here: