On the FID calculation for the COCO dataset

[mzx121]

Hello author, when conducting the evaluation, you mentioned that 10K tips from the COCO validation set were selected for image generation, may I ask how the generated images were handled in the FID calculation with the real dataset, here I used 5k data from the COCO validation set and the generated 10k data for the FID calculation, which was quite different from the calculation results given in your paper.