Home

Awesome

Similarity Guided Sampling

Source code of the CVPR 2021 paper: "3D CNNs with Adaptive Temporal Feature Resolutions".

Similarity Guided Sampling

Similarity Guided Sampling (SGS) is a differentiable module which can be plugged into existing 3D CNN architecture to reduce the computational cost (GFLOPs) while preserving the accuracy.

@inproceedings{sgs2021,
    Author    = {Mohsen Fayyaz, Emad Bahrami, Ali Diba, Mehdi Noroozi, Ehsan Adeli, Luc Van Gool, Juergen Gall},
    Title     = {{3D CNNs with Adaptive Temporal Feature Resolutions}},
    Booktitle = {{The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) }},
    Year      = {2021}
}

Installation

Please find installation instructions in INSTALL.md. You may follow the instructions in DATASET.md to prepare the datasets.

Quick Start

Follow the example in GETTING_STARTED.md.

License

The majority of this work is licensed under Apache 2.0 license. Portions of the project are available under separate license terms: SlowFast and 3D-ResNets-PyTorch.

References

The code is adapted from the following repositories:

https://github.com/facebookresearch/SlowFast

https://github.com/kenshohara/3D-ResNets-PyTorch