Code for the ECCV 2020 paper by Yiming Qian and Yasutaka Furukawa
Clone the repository:
git clone https://github.com/yi-ming-qian/interplane.gitWe use Python 3.7 and PyTorch 1.0.0 in our implementation, please install dependencies:
conda create -n interplane python=3.7
conda activate interplane
conda install pytorch==1.0.0 torchvision==0.2.1 cuda90 -c pytorch
conda install -c menpo opencv
pip install -r requirements.txtWe create our pairwise plane relationship dataset based on PlaneRCNN. Please follow the instructions in their repo to download their dataset.
Then dowload our relationship dataset from here, and do the following: (1) merge the "scans/" folder with "$ROOT_FOLDER/scans/", (2) place "contact_split/" under "$ROOT_FOLDER/", (3) place "planeae_result" under "$ROOT_FOLDER/".
We have three networks, Orientation-CNN, Contact-CNN, Segmentation-MPN, which are trained separately:
python train_angle.py train with dataset.dataFolder=$ROOT_FOLDER/python train_contact.py train with dataset.dataFolder=$ROOT_FOLDER/python train_segmentation.py train with dataset.dataFolder=$ROOT_FOLDER/Evaluate when input method is PlaneRCNN:
python predict_all.py eval with dataset.dataFolder=$ROOT_FOLDER/ resume_angle=/path/to/orientationCNN/model resume_contact=/path/to/contactCNN/model resume_seg=/path/to/segmentationMPN/model input_method=planercnnEvaluate when input method is PlaneAE:
python predict_all.py eval with dataset.dataFolder=$ROOT_FOLDER/ resume_angle=/path/to/orientationCNN/model resume_contact=/path/to/contactCNN/model resume_seg=/path/to/segmentationMPN/model input_method=planeaeTwo gpus are used for inference. The results will be saved under "experiments/predict/{RUN_ID}/results/". We also provide our pre-trained models here.
https://yi-ming-qian.github.io/
We thank the authors of PlaneRCNN and of PlaneAE. Our implementation is heavily built upon their codes.