Closed 2226450890 closed 7 months ago
The computational cost of our approach depends on the input image, as the point-query quadtree is dynamically constructed during inference. Therefore, the FLOPs also vary from image to image. You may try to use the first layer of the point-query quadtree to compute FLOPs.
Dear author,
I have tried to calculate the FLOPs for PET, but after much effort could not calculate it successfully, can you provide its calculation?
Thank you very much for your response.