Domain adaptation experiment

[bluejangbaljang]

Hi. I read your paper interestingly and have some questions.

1. The DA experiment(KITTI -> Cityscapes) that I understand is as follows, and I wonder if there's anything wrong with it's:

• training Faster RCNN as Cityscapes dataset
• After training KITTI -> Cityscapes I2I network(DUNIT), translating KITTI dataset to Cityscapes domain.
• Evaluating translated KITTI dataset with Faster RCNN

2. I wonder if the dataset used to conduct the detection performance evaluation was the KITTI testing set or the training set used for the evaluation in part.

3. You have evaluated a total of four classes(Person, Car, Truck, and Bicycle ). However, as far as I know, there is no Bicycle class in the KITTI dataset. Could you tell me how you evaluated this part?

Best,
Seokbeom