April 8, 2021

663 words 4 mins read

zju3dv/LoFTR

zju3dv/LoFTR

Code for "LoFTR: Detector-Free Local Feature Matching with Transformers", CVPR 2021

repo name zju3dv/LoFTR
repo link https://github.com/zju3dv/LoFTR
homepage https://zju3dv.github.io/loftr/
language Jupyter Notebook
size (curr.) 27168 kB
stars (curr.) 263
created 2021-03-02
license Apache License 2.0

LoFTR: Detector-Free Local Feature Matching with Transformers

Project Page | Paper

LoFTR: Detector-Free Local Feature Matching with Transformers
Jiaming Sun*, Zehong Shen*, Yu’ang Wang*, Hujun Bao, Xiaowei Zhou
CVPR 2021

demo_vid

TODO List and ETA

The entire codebase for data pre-processing, training and validation is under major refactoring and will be released around June. Please subscribe to this discussion thread if you wish to be notified of the code release. In the meanwhile, discussions about the paper are welcomed in the discussion panel.

  • Inference code and pretrained models (DS and OT) (2021-4-7)
  • Code for reproducing the test-set results (2021-4-7)
  • Webcam demo to reproduce the result shown in the GIF above (2021-4-13)
  • Training code and training data preparation (expected 2021-6-10)

Installation

# For full pytorch-lightning trainer features (recommended)
conda env create -f environment.yaml
conda activate loftr

# For the LoFTR matcher only
pip install torch einops yacs kornia

We provide the download link to

  • the scannet-1500-testset (~1GB).
  • the megadepth-1500-testset (~600MB).
  • 4 pretrained models of indoor-ds, indoor-ot, outdoor-ds and outdoor-ot (each ~45MB).

By now, the environment is all set and the LoFTR-DS model is ready to go! If you want to run LoFTR-OT, some extra steps are needed:

We use the code from SuperGluePretrainedNetwork for optimal transport. However, we can’t provide the code directly due its strict LICENSE requirements. We recommend downloading it with the following command instead.

cd src/loftr/utils  
wget https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/master/models/superglue.py 

Run the code

Match image pairs with LoFTR

from src.loftr import LoFTR, default_cfg

# Initialize LoFTR
matcher = LoFTR(config=default_cfg)
matcher.load_state_dict(torch.load("weights/indoor_ds.ckpt")['state_dict'])
matcher = matcher.eval().cuda()

# Inference
with torch.no_grad():
    matcher(batch)    # batch = {'image0': img0, 'image1': img1}
    mkpts0 = batch['mkpts0_f'].cpu().numpy()
    mkpts1 = batch['mkpts1_f'].cpu().numpy()

An example is given in notebooks/demo_single_pair.ipynb.

Online demo

Run the online demo with a webcam to reproduce the result shown in the GIF above.

cd demo
./run_demo.sh
#!/bin/bash
set -e
# set -x

if [ ! -f utils.py ]; then
    echo "Downloading utils.py from the SuperGlue repo."
    echo "We cannot provide this file directly due to its strict licence."
    wget https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/master/models/utils.py
fi

# Use webcam 0 as input source. 
input=0
# or use a pre-recorded video given the path.
# input=/home/sunjiaming/Downloads/scannet_test/$scene_name.mp4

# Toggle indoor/outdoor model here.
model_ckpt=../weights/indoor_ds.ckpt
# model_ckpt=../weights/outdoor_ds.ckpt

# Optionally assign the GPU ID.
# export CUDA_VISIBLE_DEVICES=0

echo "Running LoFTR demo.."
eval "$(conda shell.bash hook)"
conda activate loftr
python demo_loftr.py --weight $model_ckpt --input $input
# To save the input video and output match visualizations.
# python demo_loftr.py --weight $model_ckpt --input $input --save_video --save_input

# Running on remote GPU servers with no GUI.
# Save images first.
# python demo_loftr.py --weight $model_ckpt --input $input --no_display --output_dir="./demo_images/"
# Then convert them to a video.
# ffmpeg -framerate 15 -pattern_type glob -i '*.png' -c:v libx264 -r 30 -pix_fmt yuv420p out.mp4

Reproduce the testing results with pytorch-lightning

conda activate loftr
# with shell script
bash ./scripts/reproduce_test/indoor_ds.sh

# or
python test.py configs/data/scannet_test_1500.py configs/loftr/loftr_ds.py --ckpt_path weights/indoor_ds.ckpt --profiler_name inference --gpus=1 --accelerator="ddp"

For visualizing the results, please refer to notebooks/visualize_dump_results.ipynb.

Citation

If you find this code useful for your research, please use the following BibTeX entry.

@article{sun2021loftr,
  title={{LoFTR}: Detector-Free Local Feature Matching with Transformers},
  author={Sun, Jiaming and Shen, Zehong and Wang, Yuang and Bao, Hujun and Zhou, Xiaowei},
  journal={{CVPR}},
  year={2021}
}

This work is affiliated with ZJU-SenseTime Joint Lab of 3D Vision, and its intellectual property belongs to SenseTime Group Ltd.

Copyright SenseTime. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
comments powered by Disqus