In the pouring dataset /nerf_synthetic/test, there's a mix of r_x.png and r_x_flow.png (where x is the index), should the r_x_flow.png be interpreted as the optical flow map? How are they generated? Where can I find a script to generate them? In the Gibson dataset, /gibson_dataset/gibson_test_100_smaller_range_0_6_0_0_0_4, there are 5 kinds of data: d_x_flow.npy, d_x.npy, flow_x.npy, r_x.png, r_x_flow.png. Again how to interpret them and how to compute them based on the raw data (which I assume are RGB and depth images)? Where can I find a script to generate them? Also what is the data in the render_linear folder?

TL; DR: What is the structure of each dataset? How to interpret the files? Are there scripts to generate them? Thanks in advance!

How to input parameters for testing after training the radiation model? And why is the training of flow model so slow? Training the flow model on 3090Ti only ran less than 10000 steps a day.

hello，I have started training.But there is a problem, the model training is very slow. The first command ran for 20 hours, and the second stream consistency training also ran for more than 20 hours to complete 50% of the progress, and my GPU utilization was only 3%. My configuration is a single Nvidia 3090 Ti GPU. I didn't modify your code, have you ever encountered this situation?

I feel sorry to bother you again. I just finish the first phase of training with your sincerely help and move on to the next command which is python run_nerf.py --config=configs/pour_baseline_flow.txt and got FileNotFoundError: No such file: '/nerflow/data/nerf_synthetic/pouring_dataset/depth_train/r_1_flow0068.hdr'. I look into transforms_train.json and find out that item "depth_train_path" seems give the filename that doesn't match with the filename in depth_train folder actually. I just add some screenshot in case I didn't describe the problem clearly.