Question about the inference

[masterwu2115]

Hi authors, thanks for sharing your code. However, I'm confused about how to use your provided pre-trained model to generate full-resolution reconstructed images, that is, a video clip with the dimension of 8 times 3 times 512 times 512. It seems like your implementation sample 625 ray_idx rather than rays for the entire grid. Could you provide instructions on generating rendered images to reproduce results as you listed in Table 1?

[yunhaoli2020]

You can test our code and render retrieved images from compressed image with the provided checkpoints. To do this, you should first download the checkpoints, then put the weight file under the corresponding logs folder ./logs/cozy2room, and then change the parameter load_weights=True in cozy2room.txt, finally run

python train_cacti3.py --config configs/cozy2room.txt

Here I use cozy2room for example. You can switch to other scene names for other names. In this case, the program will engage testing mode and render images from pre-trained NeRF.

I hope that helps you.