Canhui Tang

感谢上次您的及时回复！关于使用测试集我又有一个疑问。 您在Table 8 中复现了DRAEM在MvTec3d数据集上的表现，仅得到75.7% I-AUROC。 然而在另一篇基于DRAEM的工作EasyNet [https://arxiv.org/pdf/2307.13925.pdf] 中，它在MvTec3d上的RGB效果达到了90.4%，包括我自己的复现结果也是90%左右。 我怀疑原因是这样的：因为DRAEM代码中没有使用测试集来选择最好的epoch，而是采用了最后一个epoch. 您可能也是这样？而EasyNet和我的复现使用了测试集做筛选。 实际上，我觉得DRAEM这种合成噪声的方法非常不稳定，它每个epoch都会产生不同的噪声，导致不同epoch模型的效果有很大差异，只有使用测试集去选模型才能有一个好的效果。

Hi, are you performing inference or training? If you are only using the weights for inference, you don't need to run scripts/merge_weights.py — that script is only required if you're...

Thank you for your attention！You may check the following: 1. Has each piece of data been tested successfully? 2. Is the correct checkpoint loaded? 3. You can run demo.py to...

Hi, thank you for your recognition. I think that R1-Multimodal-Journey does not have the bug mentioned in this blog post, because the inputs for its ref_model and current model are...