search

ethz-asl / hfnet

From Coarse to Fine: Robust Hierarchical Localization at Large Scale with HF-Net (https://arxiv.org/abs/1812.03506)
MIT License
776 stars 185 forks source link

Extremely large size npz files during SuperPoint export_predictions #65

Open Gpetrak opened 2 years ago

Gpetrak commented 2 years ago

Hi developers,

I'm trying to train Hfnet. During distillation, the exported npz files exceeded the 100 Gb in 20 minutes. As a result the process stopped because no space left on my device. My dataset is about 5 Gb of 50 000 images from Berkley DeepDrive dataset. The command that I'm using is the following:

python hfnet/hfnet/export_predictions.py hfnet/hfnet/configs/superpoint_export_distill.yaml superpoint_predictions --keys local_descriptor_map,dense_scores --as_dataset

Is it reasonable or am I doing something wrong ?