Skip to content

The official implementation of “Segment Anything Model is a Good Teacher for Local Feature Learning”.

Notifications You must be signed in to change notification settings

vignywang/SAMFeat

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

44 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SAMFeat (Local features detection and description)

Implementation of "Segment Anything Model is a Good Teacher for Local Feature Learning" (http://arxiv.org/abs/2309.16992).

Keywords: Local features detection and description; local descriptors; image matching; Segment Aything Model.

To do:

  • Evaluation code and Trained model for SAMFeat
  • Training code (Coming soon)

Requirement

conda env create -f environment.yml,

Quick start

HPatches Image Matching Benchmark

  1. Download trained SAMFeat model:

cd ckpt

Use the link https://drive.google.com/file/d/1NTRGZ2aJnT59_6b-n_SFY33jOwwxCJOM/view?usp=drive_link to download our trained model checkpoint from Google Drive. Place it under the ckpt folder.

  1. Download HPatches benchmark:

cd evaluation_hpatch/hpatches_sequences then bash download.sh

  1. configure evaluation file:

Edit SAMFeat_eva.yaml file located in the configs folder

  1. Extract local descriptors:
cd evaluation_hpatch
python export.py --top-k 10000 --tag SAMFeat --output_root output_path --config PATH_TO_SAMFeat_eva.yaml

This will extract descriptors and place it under the output folder

  1. Evaluation
python get_score.py

This will print out the MMA score from threshold 1-to-10 and output a Pdf MMA Curve

About

The official implementation of “Segment Anything Model is a Good Teacher for Local Feature Learning”.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published