But what if you’re good at it?
The standard train/test split contains strong semantic clustering and substantial similarity between training and test images within the same MS COCO categories. This can inflate apparent reconstruction performance (see Shirakawa et al., 2025). For reconstruction studies it is advisable to create alternative splits where test stimuli contain categories not present during training, or where semantic and visual overlap between training and test images is minimized.
，这一点在PDF资料中也有详细论述
李可佳：我相信同样的社区未来在其他地方一定也会有，但社区的整体水平和信息的丰富度一定会有差异。比如哈佛大学的藏书和整个氛围，基准肯定要比任意一个不知名的大学要好很多。
not yet reached a 95% confidence level given the natural temperature