myshelter-c / MFFENet

MFFENet

Preparation

The repository was built in PyTorch 1.7.1 and trained and tested on the environment (Python 3.7, CUDA 11.6).

Preparing the environment:

conda create -n name python=3.7
conda activate name

Install PyTorch dependencies

pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

Datasets

Please refer to the data preparation method of monodepth2.

Training

We provide training using HRFormer pre-trained weights. You can download it and place it in the models folder.
Our model runs on two 16GB GPUs, and we support distributed training.

python -u -m torch.distributed.launch --nproc_per_node=2 train.py --model_name M_640x192_name --png --batch_size 8 --data_path data_path/kitti

Testing

python evaluate_depth.py   --load_weights_folder models/M_640x192_name --eval_mono --eval_split eigen --data_path data_path/kitti

Weights trained on the KITTI dataset

We provide weights for:

• MFFENet-tiny-model(640x192)
• MFFENet-small-model(640x192)

Acknowledgement

Thanks to the authors for their excellent work:

• HRFormer
• DIFFNet
• monodepth2