Hello, first of all thank you for your very nice work !
I've had a question concerning the computing of FID score on COCO dataset as detailed in section 5.2. I've not found inside the repo how you would do it, my question is how did you compute this metric ?
Is it directly on the slots from input images of COCO ? I mean by that you extract slots and then direclty reconstruct input images ? Or did you apply the procedure like in https://arxiv.org/pdf/2303.10834 (Object-centric Slot Diffusion) where you first cluster concepts and then randomly pick slots to regenerate images ? Then, if it's the second procedure, how would you compute the other metrics (PSNR, SSIM, LPIPS) as you don't have input image to compare ?
Thank you in advance
Hello, first of all thank you for your very nice work !
I've had a question concerning the computing of FID score on COCO dataset as detailed in section 5.2. I've not found inside the repo how you would do it, my question is how did you compute this metric ?
Is it directly on the slots from input images of COCO ? I mean by that you extract slots and then direclty reconstruct input images ? Or did you apply the procedure like in https://arxiv.org/pdf/2303.10834 (Object-centric Slot Diffusion) where you first cluster concepts and then randomly pick slots to regenerate images ? Then, if it's the second procedure, how would you compute the other metrics (PSNR, SSIM, LPIPS) as you don't have input image to compare ?
Thank you in advance