Experiments on real datasets

[chen8573]

Hello author, I have two questions about experiments on real datasets:

1.In the comparative experiment section, for the results on the real set, did you only use the real training set (5051 images) for training in other methods?

2. In the ablation experiment section, before using SJDL, did the baseline and JDL experiments only use the real training set (5051 images) for training the results on the real set? ![Uploading image.png…]()

[chen8573]

comparative experiment： ablation experiment ：

[AaronCIH]

@chen8573 Hi, for the purpose of a fair comparison, method (original) is trained on a clear dataset method-defog is trained on a clear dataset, but the defoggy image is used in the inference stage. method-fog is directly trained on a combined FVRID-syn and FVRID-real training set. all of our methods are alternately trained on the FVRID-syn and FVRID-real training set.

[chen8573]

In the JDL experiment section, does alternating training mean that the images from the real training set only go through the re recognition branch without going through the defogging branch?