Training with multi-task distillation

Exporting the target predictions

We first export the predictions of NetVLAD (global descriptor) and SuperPoint (dense keypoint scores and descriptors), which will be the labels of the dataset.

python3 hfnet/export_predictions.py
	hfnet/configs/netvlad_export_distill.yaml \
	global_descriptors \
	--keys global_descriptor \
	--as_dataset
python3 hfnet/export_predictions.py
	hfnet/configs/superpoint_export_distill.yaml \
	superpoint_predictions \
	--keys local_descriptor_map,dense_scores \
	--as_dataset

Training HF-Net

python3 hfnet/train.py hfnet/configs/hfnet_train_distill.yaml hfnet

The training can be interrupted at any time using Ctrl+C and can be monitored with Tensorboard summaries saved in $EXPER_PATH/hfnet/. The weights are also saved there.

Exporting the model for deployment

python3 hfnet/export_model.py config/hfnet_train_distill.yaml hfnet

will export the model to $EXPER_PATH/saved_models/hfnet/.