Train and Test

[asher-bit]

Hello, I try to train and test your work with what you say in readme.md. But i have a question that why train and test are the same dataset(chari.txt) And Why there is no dataset_path in compress progress.

[asher-bit]

The three images are training, encoding and decoding.

[daniel03c1]

First of all, sorry for the late response.

The reason for using the same dataset for training and testing is that NeRF requires each network for each dataset. If you are talking about the train and test split, the training code automatically splits images into separate sets.

The reason why a datapath is not required in the compression process is that our compressing process does not require any tuning during the compression process.

[asher-bit]

Hello, I have another question for the paper. After using tensoRF decomposition to get three sets of planes and vectors, why use idwt instead of dwt. In addition, where is the wavelet coefficient in Figure 2 obtained in the paper? Looking forward to your response.

[daniel03c1]

The reason for using idwt is that we view planes as wavelet coefficients. Using iDWT, we can project these wavelet coefficients into the spatial domain for further rendering. Therefore, the planes used for training are exactly wavelet coefficients. I hope I have addressed your concerns well.

[asher-bit]

Can I understand that by adding an inverse wavelet transform, the coefficients obtained when decomposing the three-dimensional mesh into planes can be more sparse (closer to the coefficients obtained with the wavelet transform)

[daniel03c1]

If wavelet coefficients were to be used, they must be inverse transformed before rendering. That is why we used iDWT, so they are not close to wavelet coefficients, but they are the wavelet coefficients.

[asher-bit]

OkI think i know your means. I try to not use dwt when train with "python3 train.py --config=configs/chair.txt --use_mask --mask_weight=1e-10 --grid_bit=8" and use the same compress and decompress process. But i get PSNR with 14.xxx. Is my command wrong or is there something else?

[daniel03c1]

Could you please check if you used the latest version by any chance?

[asher-bit]

I use this branch for test.

[asher-bit]

I solved this problem by commenting out the code using inverse in compress. Thank you for your reply and your great work!