SpaceJAM: a Lightweight and Regularization-free Method for Fast Joint Alignment of Images (ECCV 2024)
Nir Barel*, Ron Shapira Weber*, Nir Mualem, Shahaf Finder and Oren Ferifeld
Our framework jointly aligns a set of images of an object category in only a few minutes.
Top-to-bottom: 1) input images; 2) learned low-dimensional representations; 3) aligned features; 4) aligned images.
The last column depicts the average representation (atlas) obtained after training.
To set up the environment for this project, you need to install the required dependencies listed in environment.yml
. This file specifies the necessary packages and channels to ensure that your environment is properly configured.
- Install Conda: If you don't have Conda installed, you can get it by installing Miniconda or Anaconda. Download Miniconda or Download Anaconda.
- Create the Environment: To create a Conda environment with the dependencies specified in
environment.yml
, use the following command:
conda env create -f environment.yml
- Activate the Environment: Once the environment is created, activate it using:
conda activate spacejam
You can download and preprocess the following datasets according to the paper:
python prepare_data/prepare_spair.py
python prepare_data/prepare_cub_class.py [--cub_acsm_class <class_num>]
python prepare_data/prepare_cub_subsets.py
In order to run SpaceJAM on custom images set, use:
python prepare_data/prepare_image_set.py --path <image-dir> [--out <out-dir>]
To train the entire model on one of the dataset, simply run:
python train.py --data_folder <processed-images-dir>
For more details, run:
python train.py --help
@inproceedings{barel2024spacejam,
title={SpaceJAM: a Lightweight and Regularization-free Method for Fast Joint Alignment of Images},
author={Nir Barel and Ron Shapira Weber and Nir Mualem and Shahaf E. Finder and Oren Freifeld},
year={2024},
eprint={2407.11850},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2407.11850},
}