How do some datasets perform inference without a training set?

[CheneyXuYiMin]

In your paper, there are no training sets for CHEMLEON, NC4K and other datasets. Is it zero-sample reasoning? Or use training models from other datasets to reason about them? Looking forward to your answer, thank you very much!

[xiongxyowo]

Hi, the training set contains 4040 (1000 CAMO + 3040 COD10K) images in total. Once trained on the joint training set, the model can be tested on the four test sets. It is recommended to refer to the experimental sections of other COD papers to get familiar with the dataset configurations of the COD benchmark.

[CheneyXuYiMin]

So in Salient Object Detection, is the same for HKU-IS and DUT-OMRON?

[xiongxyowo]

Yes

[CheneyXuYiMin]

Yes

Thanks for your response