Hi, does anyone have an example they can share for a PerceiverIO model working with position embeddings? For instance image segmentation. I am using the fourier_encode method to create position embeddings which I a) append to the input images and b) use as query vectors in order to get my output. But the network does train and instead produces some stripy-patterns as an output
Hi, does anyone have an example they can share for a PerceiverIO model working with position embeddings? For instance image segmentation. I am using the
fourier_encodemethod to create position embeddings which I a) append to the input images and b) use as query vectors in order to get my output. But the network does train and instead produces some stripy-patterns as an output