Questions while using pre-trained model on custom data

[orpheus92]

I have a couple of questions while using pre-trained model to synthesize scenes for custom dataset.
First of all, I was able to generate and render results within the boundary. However, I noticed that some furnitures were not oriented correctly. It does not help if I remove the cosin_angle in the config file.
Second, I noticed that when you render ground truth data for 3D Front via render_threedfront_scene.py, you can use "-with_door_and_windows". However, I did not find information related to doors and windows for custom input. How are windows and doors treated for test data not in 3D Front?
Third, besides render_results that give a top-down projection of the room, are there other ways to visualize the room layout for custom test dataset? Like visualizations used in the paper?
Finally, regarding the training resources and benchmarks described in the paper, do you think RTX 3080 should also work for the training on custom dataset?