search

Parskatt / DeDoDe

[3DV 2024 Oral] DeDoDe 🎶 Detect, Don't Describe --- Describe, Don't Detect, for Local Feature Matching
MIT License
367 stars 19 forks source link

How to train on my data #20

Open GavinYang5 opened 10 months ago

GavinYang5 commented 10 months ago

Hi, your job is fantastic! I have RGB and depth information collected by an RGBD camera, and I can obtain the camera pose through the calibration plate. Do I already have all the data needed for training? How should I prepare the data?

Parskatt commented 10 months ago

Hi, then it should be possible to train the descriptor part. The detector relies on SfM, so can be a bit tricky to retrain.

You will need to make a dataloader for you dataset that gives you the two images, depth, relative pose, and intrinsics in a similar way as megadepth.py.

Let me know if you face any issues :)

GavinYang5 commented 10 months ago

Thank you for your reply. I am not a person engaged in this field. I just want to estimate the pose information of the object through RGB images, using ransac pnp. I think I should first understand the structure of the megadepth data set and try to use my data to build a data set with the same structure as megadepth