TRI-ML/KP2D

How to get Qualatitive Results?

UmutCosk opened this issue · 4 comments

I would like to have the qualitative results like in the readme for image pairs.

How would I set it up?

Hi, for drawing keypoints on an images we have this function in the repo. For drawing corrrespondences you could use something like opencv's drawMatches.

How can I get the keypoints and descriptors using the pre-trained models?

Try running:

make docker-run COMMAND="python scripts/eval_keypoint_net.py --pretrained_model /data/models/kp2d/v4.ckpt --input /data/datasets/kp2d/HPatches/"

You can then get the scores, keypoints and descriptors here.

Hi everyone,
has someone implemented it yet ? I would have liked to take a look at it before doing it myself.