DmifNet

Citing this work

If you find our code or paper useful, please consider citing

@inproceedings{li2021dmifnet,
  title={DmifNet: 3D Shape Reconstruction based on Dynamic Multi-Branch Information Fusion},
  author={Li, Lei and Wu, Suping},
  booktitle={2020 25th International Conference on Pattern Recognition (ICPR)},
  pages={7219--7225},
  year={2021},
  organization={IEEE}
}

Configuration Environment

Python3.6, Pytorch: 1.0, CUDA: 9.0+, Cudnn: follow Cuda version, GPU: one Nvidia RTX 2080Ti
Epoch: Approximately close to the optimal convergence value between 1500 and 2500 epoch

Installation

First you have to make sure that you have all dependencies in place.

You can create an anaconda environment called dmifnet_space using

conda env create -f dmif_env.yaml
conda activate dmifnet_space

Then, compile the extension modules.

python set_env_up.py build_ext --inplace

Then, download the BatchNet modules.

• download the BatchNet via BaiDu and Extracted key is 1234
• download the BatchNet via Google

Next, put your model path in DmifNet/dmifnet/encoder/batchnet.py /def resnet18(pretrained=False, **kwargs):

Generation

To generate meshes using a trained model, use

python generate.py ./config/img/dmifnet.yaml

Training

python train.py ./config/img/dmifnet.yaml

DataSet

You can check the baseline work Onet to download the datasetONet and DmifNet: DataSet. Thanks for contribution of baseline work.

Evaluation

First, to generate meshes using a trained model, use

python generate.py ./config/img/dmifnet.yaml

Then, for evaluation of the models, you can run it using

python eval_meshes.py ./config/img/dmifnet.yaml

also can use quick evaluation(don't need generation).

python eval.py ./config/img/dmifnet.yaml

Pretrained model

you can download our pretrained model via Baidu Netdisk or Google Drive

• download the DmifNet via BaiDu and Extracted key is 3hfs
• download the DmifNet via Google

Quantitative Results

Qualitative Results

Futher Information

If you have any problems with the code, please list the problems you encountered in the issue area, and I will reply you soon. Thanks for baseline work Occupancy Networks - Learning 3D Reconstruction in Function Space.