An appropriate metric for evaluating the similarity of targets with potentially different sizes and positions ？

[comrade bionic]

Awesome work ! Thank you for sharing. I'd like to ask about comparing two images: one being the ground truth image and the other the reconstructed result. Due to certain inherent properties, the targets in the reconstructed image are structurally similar to those in the original, but may differ in size and position. If positional accuracy isn't a concern and the primary focus is on assessing target similarity, using SSIM directly might not be the most appropriate. I'm considering using LPIPS in this case. Could you advise on which metric would be more suitable for this scenario?