This is the official repository of the paper Memory Scalable and Simplified Functional Map Learning, published at CVPR 2024
This is a v0.1 of the codebase which should be able to reproduce the results exactly. There is no explicit dependency on ScalableDenseMaps in this version, an early versio of the package is provided in the DiffZo folder.
A simpler and cleaner version of the codebase will be released soon, relying cleanly on ScalableDenseMaps.
The code is essentially the one from AttentiveFMaps, but providing a new trainer, and adding things related to Differentiable ZoomOut.
To train the model, use
python trainer_dzo.py run_mode=train run_cfg=path/to/cfg/config.ymlThe data can be found on the AttentiveFMaps github.
You can download pretrained models at this address. You can then run evaluation using this code:
python trainer_dzo.py run_mode=test run_ckpt=path/to/ckpt/ckpt.pthYou can run your own evaluation using the pretrained checkpoints, where the DiffusionNet's weights are stored in the feature_extractor key.
If you use this work, please cite
@inproceedings{magnetMemoryScalable2024,
title = {Memory Scalable and Simplified Functional Map Learning},
booktitle = {2024 {{IEEE}}/{{CVF Conference}} on {{Computer Vision}} and {{Pattern Recognition}} ({{CVPR}})},
author = {Magnet, Robin and Ovsjanikov, Maks},
year = {2024},
publisher = {IEEE},
}