SuperPoint-PrP

This project is part of the codebase for NeRF-Supervised Feature Point Detection and Description.

The SuperPoint-PrP implementation is based on:

• The SuperPoint paper, Tensorflow implementation and GlueFactory's implementation.

1. Setup

In order to install the requirements and setup SuperPoint-PrP and the paths, run:

make install

You will be required to provide three different paths:

1. Data_PATH: The path to the folder which will contain the datasets.
2. CKPT_PATH: The path where the model's checkpoints are saved.
3. EXPER_PATH: The path of the directory where experiments are written.

The Dataset can be downloaded through the following link.

The folder containing the datasets should be structured as follows:

| datasets
|   |-- NeRF
|   |  |-- images
|   |  |   |-- training
|   |  |   |-- validation
|   |  |-- camera_transforms
|   |  |   |-- training
|   |  |   |-- validation
|   |  |-- depth
|   |  |   |-- training
|   |  |   |-- validation
|   |-- HPatches
|   |   |-- i_ajustment
|   |   |   |--1.ppm
|   |   |   |--...
|   |   |   |--H_1_2
|   |   |-- ...
|   |-- ScanNet
|   |   |-- ....

2. Configurations

To display all available training options run:

python engine.py -h

3. Training SuperPoint-PrP

python engine.py --config_path .\configs\train.yaml --task train

4. Evaluating HPatches Homography Estimation and Repeatability

Run the following to run the HPatches evaluation:

python engine.py --config_path .\configs\hpatches_eval.yaml --task hpatches_evaluation

In the configuration file, change the alteration argument as v to evaluate using varying viewpoint scenes only, i to evaluate on varying illumination scenes, or all to run full Hpatches evaluation.

5. Relative Pose Estimation Evaluation

Run the following to run the relative pose estimation on the ScanNet dataset:

python engine.py --config_path .\configs\scannet_pose.yaml --task pose_evaluation

For YFCC outdoor relative pose estimation evaluation, run the following:

python engine.py --config_path .\configs\YFCC_pose.yaml --task pose_evaluation

Credits

Special thanks to Daniel DeTone, Tomasz Malisiewicz, Andrew Rabinovich the authors of SuperPoint, Rémi Pautrat for the Tensorflow implementation and the authors of Superpoint's GlueFactory implementation.