https://github.com/vignywang/SAMFeat

The official implementation of “Segment Anything Model is a Good Teacher for Local Feature Learning”.
https://github.com/vignywang/SAMFeat

Last synced: about 2 months ago
JSON representation

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

• Host: GitHub
• URL: https://github.com/vignywang/SAMFeat
• Owner: vignywang
• Created: 2023-09-13T09:23:07.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-10-07T09:12:03.000Z (over 1 year ago)
• Last Synced: 2023-10-07T12:56:04.553Z (over 1 year ago)
• Language: Python
• Size: 54.7 KB
• Stars: 23
• Watchers: 5
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• Awesome-Segment-Anything - [code

README

        
# 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：

- [x] 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.

2. Download HPatches benchmark: 

```cd evaluation_hpatch/hpatches_sequences``` then ```bash download.sh```

3. configure evaluation file:

Edit ```SAMFeat_eva.yaml``` file located in the ```configs``` folder

4. 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

5. Evaluation

```

python get_score.py

```

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