About the size of dataset.

[DREAMXFAR]

Hello, I am confused of the number of images in Human-Art, and I want some help.

As it claims that Human-Art contains 50,000 images including human figures in 20 scenarios (5 natural scenarios, 3 2D artificial scenarios, and 12 2D artificial scenarios), but I only find it contains 38 ,000 (2000*19), which is consistent with the image number in training_humanart.json + validation_humanart.json. So, I wonder whether the class generated image is missed? Or Human-Art actually only contain 38k images? How many images are used to train the model in the paper?

Hope for your kind responses.