Buggy Preprocessing

[saliteta]

When try to train the affinity features, I found there is no feature preprocessing steps. And it seems like there is little introduction of preprocessing.
To run it:
Use

python preprocess.py --dataset_path {scene_location} --sam_ckpt sam_vit_h_4b8939.pth  

But it will save the feature data as npy in language_feature_test/_s.npy and /_f.npy

In train_affinity**.py it will use language_feature without _test post-fix.

Therefore, I change this line

see scene/dataset_readers_multi_lvl.py line 172
I changed

mask_dir = "language_features" to mask_dir = "language_features_test"

Is the above change make sense?

[Jumpat]

Yeah. I think it's okay!