Closed zhangwenniu closed 1 week ago
Hi to use the SemanticParser for datasets like RobustNeRF, you need to arrange your data folder like below:
data
|_ images
| |_ clean000.jpg
| ...
| |_ extra000.jpg
| ...
| |_ clutter000.jpg
| ...
|_ sparse
|_ 0
|_ cameras.bin
|_ points3D.bin
|_ images.bin
The images have prefixes like "clean" for clean data and "extra" for test clean images and "clutter" for noisy images. You can have different prefixes and pass the prefix for train and test images to the training command. For example we use "clutter" prefix for training images and "extra" for test images, in the robust training scenario. We use "clean" for training for the idea clean set up in the paper (usually this is not available for custom real data so you can ignore it).
Thanks.
Hi, thanks for your great contribution for robust reconstruction.
But I'm confused about how to train on my dataset, because I don't know exactly how RobustNeRF dataset is built.
For example, when I get jpg images in the folder of 'data/test1/img_001.jpg', what steps should I do to get proper data formats to train the model?
Looking forward to your kind reply.