Patch2Pix: Epipolar-Guided Pixel-Level Correspondences [CVPR2021]

Overview

Patch2Pix for Accurate Image Correspondence Estimation

This repository contains the Pytorch implementation of our paper accepted at CVPR2021: Patch2Pix: Epipolar-Guided Pixel-Level Correspondences. [Paper] [Video].

Overview To use our code, first download the repository:

git clone [email protected]:GrumpyZhou/patch2pix.git

Setup Running Environment

The code has been tested on Ubuntu (16.04&18.04) with Python 3.7 + Pytorch 1.7.0 + CUDA 10.2.
We recommend to use Anaconda to manage packages and reproduce the paper results. Run the following lines to automatically setup a ready environment for our code.

conda env create -f environment.yml
conda activte patch2pix

Download Pretrained Models

In order to run our examples, one needs to first download our pretrained Patch2Pix model. To further train a Patch2Pix model, one needs to download the pretrained NCNet. We provide the download links in pretrained/download.sh. To download both, one can run

cd pretrained
bash download.sh

Evaluation

❗️ NOTICE ❗️ : In this repository, we only provide examples to estimate correspondences using our Patch2Pix implemenetation.

To reproduce our evalutions on HPatches, Aachen and InLoc benchmarks, we refer you to our toolbox for image matching: image-matching-toolbox. There, you can also find implementation to reproduce the results of other state-of-the-art methods that we compared to in our paper.

Matching Examples

In our notebook examples/visualize_matches.ipynb , we give examples how to obtain matches given a pair of images using both Patch2Pix (our pretrained) and NCNet (our adapted). The example image pairs are borrowed from D2Net, one can easily replace it with your own examples.

Training

Notice the followings are necessary only if you want to train a model yourself.

Data preparation

We use MegaDepth dataset for training. To keep more data for training, we didn't split a validation set from MegaDepth. Instead we use the validation splits of PhotoTourism. The following steps describe how to prepare the same training and validation data that we used.

Preapre Training Data

  1. We preprocess MegaDepth dataset following the preprocessing steps proposed by D2Net. For details, please checkout their "Downloading and preprocessing the MegaDepth dataset" section in their github documentation.

  2. Then place the processed MegaDepth dataset under data/ folder and name it as MegaDepth_undistort (or create a symbolic link for it).

  3. One can directly download our pre-computred training pairs using our download script.

cd data_pairs
bash download.sh

In case one wants to generate pairs with different settings, we provide notebooks to generate pairs from scratch. Once you finish step 1 and 2, the training pairs can be generated using our notebook data_pairs/prep_megadepth_training_pairs.ipynb.

Preapre Validation Data

  1. Use our script to dowload and extract the subset of train and val sequences from the PhotoTourism dataset.
cd data
bash prepare_immatch_val_data.sh
  1. Precompute image pairwise overlappings for fast loading of validation pairs.
# Under the root folder: patch2pix/
python -m data_pairs.precompute_immatch_val_ovs \
		--data_root data/immatch_benchmark/val_dense

Training Examples

To train our best model:

python -m train_patch2pix --gpu 0 \
    --epochs 25 --batch 4 \
    --save_step 1 --plot_counts 20 --data_root 'data' \
    --change_stride --panc 8 --ptmax 400 \
    --pretrain 'pretrained/ncn_ivd_5ep.pth' \
    -lr 0.0005 -lrd 'multistep' 0.2 5 \
    --cls_dthres 50 5 --epi_dthres 50 5  \
    -o 'output/patch2pix' 

The above command will save the log file and checkpoints to the output folder specified by -o. Our best model was trained on a 48GB GPU. To train on a smaller GPU, e.g, with 12 GB, one can either set --batch 1 or --ptmax 250 which defines the maximum number of match proposals to be refined for each image pair. However, those changes might also decrease the training performance according to our experience. Notice, during the testing, our network only requires 12GB GPU.

Usage of Visdom Server Our training script is coded to monitor the training process using Visdom. To enable the monitoring, one needs to:

  1. Run a visdom sever on your localhost, for example:
# Feel free to change the port
python -m visdom.server -port 9333 \
-env_path ~/.visdom/patch2pix
  1. Append options -vh 'localhost' -vp 9333 to the commands of the training example above.

BibTeX

If you use our method or code in your project, please cite our paper:

@inproceedings{ZhouCVPRpatch2pix,
        author       = "Zhou, Qunjie and Sattler, Torsten and Leal-Taixe, Laura",
        title        = "Patch2Pix: Epipolar-Guided Pixel-Level Correspondences",
        booktitle    = "CVPR",
        year         = 2021,
}
Owner
Qunjie Zhou
PhD Candidate at the Dynamic Vision and Learning Group.
Qunjie Zhou
Storchastic is a PyTorch library for stochastic gradient estimation in Deep Learning

Storchastic is a PyTorch library for stochastic gradient estimation in Deep Learning

Emile van Krieken 140 Dec 30, 2022
Pytorch Lightning code guideline for conferences

Deep learning project seed Use this seed to start new deep learning / ML projects. Built in setup.py Built in requirements Examples with MNIST Badges

Pytorch Lightning 1k Jan 02, 2023
Revisiting Self-Training for Few-Shot Learning of Language Model.

SFLM This is the implementation of the paper Revisiting Self-Training for Few-Shot Learning of Language Model. SFLM is short for self-training for few

15 Nov 19, 2022
Codebase for testing whether hidden states of neural networks encode discrete structures.

structural-probes Codebase for testing whether hidden states of neural networks encode discrete structures. Based on the paper A Structural Probe for

John Hewitt 349 Dec 17, 2022
Implementation of "Efficient Regional Memory Network for Video Object Segmentation" (Xie et al., CVPR 2021).

RMNet This repository contains the source code for the paper Efficient Regional Memory Network for Video Object Segmentation. Cite this work @inprocee

Haozhe Xie 76 Dec 14, 2022
Train CPPNs as a Generative Model, using Generative Adversarial Networks and Variational Autoencoder techniques to produce high resolution images.

cppn-gan-vae tensorflow Train Compositional Pattern Producing Network as a Generative Model, using Generative Adversarial Networks and Variational Aut

hardmaru 343 Dec 29, 2022
This is the official implement of paper "ActionCLIP: A New Paradigm for Action Recognition"

This is an official pytorch implementation of ActionCLIP: A New Paradigm for Video Action Recognition [arXiv] Overview Content Prerequisites Data Prep

268 Jan 09, 2023
Extending JAX with custom C++ and CUDA code

Extending JAX with custom C++ and CUDA code This repository is meant as a tutorial demonstrating the infrastructure required to provide custom ops in

Dan Foreman-Mackey 237 Dec 23, 2022
Re-implement CycleGAN in Tensorlayer

CycleGAN_Tensorlayer Re-implement CycleGAN in TensorLayer Original CycleGAN Improved CycleGAN with resize-convolution Prerequisites: TensorLayer Tenso

89 Aug 15, 2022
Face Mesh is a face geometry solution that estimates 468 3D face landmarks in real-time even on mobile devices

Face-Mesh Face Mesh is a face geometry solution that estimates 468 3D face landmarks in real-time even on mobile devices. It employs machine learning

Farnam Javadi 9 Dec 21, 2022
Code for 'Single Image 3D Shape Retrieval via Cross-Modal Instance and Category Contrastive Learning', ICCV 2021

CMIC-Retrieval Code for Single Image 3D Shape Retrieval via Cross-Modal Instance and Category Contrastive Learning. ICCV 2021. Introduction In this wo

42 Nov 17, 2022
Source code for the plant extraction workflow introduced in the paper “Agricultural Plant Cataloging and Establishment of a Data Framework from UAV-based Crop Images by Computer Vision”

Plant extraction workflow Source code for the plant extraction workflow introduced in the paper "Agricultural Plant Cataloging and Establishment of a

Maurice Günder 0 Apr 22, 2022
Learning Versatile Neural Architectures by Propagating Network Codes

Learning Versatile Neural Architectures by Propagating Network Codes Mingyu Ding, Yuqi Huo, Haoyu Lu, Linjie Yang, Zhe Wang, Zhiwu Lu, Jingdong Wang,

Mingyu Ding 36 Dec 06, 2022
CT-Net: Channel Tensorization Network for Video Classification

[ICLR2021] CT-Net: Channel Tensorization Network for Video Classification @inproceedings{ li2021ctnet, title={{\{}CT{\}}-Net: Channel Tensorization Ne

33 Nov 15, 2022
An implementation of a sequence to sequence neural network using an encoder-decoder

Keras implementation of a sequence to sequence model for time series prediction using an encoder-decoder architecture. I created this post to share a

Luke Tonin 195 Dec 17, 2022
[SIGGRAPH Asia 2019] Artistic Glyph Image Synthesis via One-Stage Few-Shot Learning

AGIS-Net Introduction This is the official PyTorch implementation of the Artistic Glyph Image Synthesis via One-Stage Few-Shot Learning. paper | suppl

Yue Gao 102 Jan 02, 2023
Build upon neural radiance fields to create a scene-specific implicit 3D semantic representation, Semantic-NeRF

Semantic-NeRF: Semantic Neural Radiance Fields Project Page | Video | Paper | Data In-Place Scene Labelling and Understanding with Implicit Scene Repr

Shuaifeng Zhi 243 Jan 07, 2023
RE3: State Entropy Maximization with Random Encoders for Efficient Exploration

State Entropy Maximization with Random Encoders for Efficient Exploration (RE3) (ICML 2021) Code for State Entropy Maximization with Random Encoders f

Younggyo Seo 47 Nov 29, 2022
PyTorch implementation for NED. It can be used to manipulate the facial emotions of actors in videos based on emotion labels or reference styles.

Neural Emotion Director (NED) - Official Pytorch Implementation Example video of facial emotion manipulation while retaining the original mouth motion

Foivos Paraperas 89 Dec 23, 2022
🏃‍♀️ A curated list about human motion capture, analysis and synthesis.

Awesome Human Motion 🏃‍♀️ A curated list about human motion capture, analysis and synthesis. Contents Introduction Human Models Datasets Data Process

Dennis Wittchen 274 Dec 14, 2022