Open ChenYutongTHU opened 5 months ago
In fact, I found for subject 313, there are only 21 cameras. Camera (20)(21) are missing. For subjects 387 and 377, there are 23 cameras. Another distinction from the paper is that for subject 377, the WarmUp sequence, in its config.yaml, cfg.num_eval_frame=200.
However, in the paper's table, the number of test frames is 317.
Which test set did you use to report the number in the table? Great thank!
In the TPAMI version, we use this config: https://github.com/zju3dv/animatable_nerf/blob/master/configs/aligned_nerf_pdf/aligned_aninerf_pdf_377.yaml#L28, which will test all the remaining frames with an interval, resulting in 317 test frames.
I see. So, there are 617 frames in total, (I checked the dataset). I guess with 300 frames for training, there are 317 remaining frames for testing, which are then sampled with interval=30.
Then how about the number of cameras?
Please see the config file. I use four cameras for training and remaining cameras for test.
Hi,
The paper mentions that 'using a multi-camera system that has 21 synchronized cameras. We select four cameras for training and use the remaining cameras for testing.'
However, I checked the config.yaml. For ZJU dataset, cfg.training_view is set [0,6,12,18] And there are 23 cameras in total. So the remaining views for testing are [1,2,3,4,5,7,8,9,10,11,13,14,15,16,17,19,20,21,22]
https://github.com/zju3dv/animatable_nerf/blob/f9b29c54b334e2d8404ede6a5badab6bc747dd22/lib/config/config.py#L24 https://github.com/zju3dv/animatable_nerf/blob/f9b29c54b334e2d8404ede6a5badab6bc747dd22/lib/datasets/tpose_pdf_dataset.py#L26-L33
Thank you!