github.com/ashutosh1807/PixelFormer @main

PixelFormer: Attention Attention Everywhere: Monocular Depth Prediction with Skip Attention

This is the official PyTorch implementation for WACV 2023 paper 'Attention Attention Everywhere: Monocular Depth Prediction with Skip Attention'.

Paper

Installation

conda create -n pixelformer python=3.8
conda activate pixelformer
conda install pytorch=1.10.0 torchvision cudatoolkit=11.1
pip install matplotlib tqdm tensorboardX timm mmcv

Datasets

You can prepare the datasets KITTI and NYUv2 according to here, and then modify the data path in the config files to your dataset locations.

Training

First download the pretrained encoder backbone from here, and then modify the pretrain path in the config files.

Training the NYUv2 model:

python pixelformer/train.py configs/arguments_train_nyu.txt

Training the KITTI model:

python pixelformer/train.py configs/arguments_train_kittieigen.txt

Evaluation

Evaluate the NYUv2 model:

python pixelformer/eval.py configs/arguments_eval_nyu.txt

Evaluate the KITTI model:

python pixelformer/eval.py configs/arguments_eval_kittieigen.txt

Pretrained Models

• You can download the pretrained models "nyu.pt" and "kitti.pt" from here.

Citation

If you find our work useful in your research, please cite the following:

@InProceedings{Agarwal_2023_WACV,
    author    = {Agarwal, Ashutosh and Arora, Chetan},
    title     = {Attention Attention Everywhere: Monocular Depth Prediction With Skip Attention},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2023},
    pages     = {5861-5870}
}

Contact

For questions about our paper or code, please contact (@ashutosh1807) or raise an issue on GitHub.

Acknowledgements

Most of the code has been adpated from CVPR 2022 paper NewCRFS. We thank Weihao Yuan for releasing the source code for the same.

Also, thanks to Microsoft Research Asia for opening source of the excellent work Swin Transformer.