Skip to content

Latest commit

 

History

History
144 lines (109 loc) · 5.57 KB

README.md

File metadata and controls

144 lines (109 loc) · 5.57 KB
Raw

PanoDR: Spherical Panorama Diminished Reality for Indoor Scenes.

Paper Paper Conference Workshop YouTube Project Page

Model Architecture


Prerequisites

  • Windows10 or Linux
  • Python 3.7
  • NVIDIA GPU + CUDA CuDNN
  • PyTorch 1.7.1 (or higher)

Installation

  • Clone this repo:
git clone https://github.com/VCL3D/PanoDR.git
cd PanoDR
  • We recommend setting up a virtual environment (follow the virtualenv documentation). Once your environment is set up and activated, install the vcl3datlantis package:
cd src/utils
pip install -e .

Dataset

We use Structured3D dataset. To train a model on the dataset please download the dataset from the official website. We follow the official training, validation, and testing splits as defined by the authors. After downloading the dataset, split the scenes into training train, validation and test folders. The folders should have the following format:

Structured3D/
    train/
        scene_00000
        scene_00001
        ...
    validation/
        scene_03000
        scene_03001
        ...
    test/
        scene_03250
        scene_03251
        ...

In order to estimate the dense layout maps, specify the path to train and test folders and run:

python src\utils\vcl3datlantis\dataset\precompute_structure_semantics.py

Training

In order to train the model, first specify the required parameters:

  • --train_path : /../Structured3D/train/
  • --test_path : /../Structured3D/test/
  • --results_path : The folder where metrics are saved
  • --gt_results_path : The folder where ground truth images are saved for testing
  • --pred_results_path : The folder where predicted images are saved for testing
  • --segmentation_model_chkpnt : The path for the pre-trained dense layout estimation model
  • --model_folder : The folder where checkpoints are saved

After starting visdom on ther server:

python -m visdom

run:

python src/train/train.py --visdom

Inference

You can download the pre-trained models from here and specify the arguments --eval_chkpnt_folder and --segmentation_model_chkpnt, respectively. Assuming the input image and mask are in the format as in the input folder run:

python src/train/test.py --inference --eval_path input/

Model service

Model is also available via torchserve. First, install the required dependencies via

cd service
pip install -r requirements.txt

Next, download the .mar file from here and place it under service/model_store. In order to serve the model using REST calls, run:

torchserve --start --ncs --model-store ./model_store --models panodr=/model_store/panodr.mar torchserve --start --ncs --model-store ./model_store --models panodr=/model_store/panodr.mar

Once the model is served, the endpoint is reachable on http://IP:8080/predictions/panodr, with IP as selected when configuring torchserve (typically localhost, but more advanced configuration is also possible to serve the model externally or make it reachable from other machines, using the inference_address setting).

A server is provided for hosting inputs and saving the output files. It can be started via:

cd Imageserver\ 
python .\imageserver.py

All images are hosted on http://IP:PORT. Further, an endpoint on http://IP:PORT/save/inpainted is provided for obtaining the output files from the service.

The following arguments have to be specified in inputs/requests.json file to call the service:

  • DataInputs["rgb"]
  • DataInputs["mask"]

Finally, to obtain predictions from the model, a callback URL json payload needs to be POSTed. Simply run:

curl.exe -X POST http://IP:8080/predictions/panodr -H "Content-Type: application/json" -d @/PATH_TO/PanoDR/service/inputs/request.json

Citation

If you use this code for your research, please cite the following:

@inproceedings{gkitsas2021panodr,
  title={PanoDR: Spherical Panorama Diminished Reality for Indoor Scenes},
  author={Gkitsas, Vasileios and Sterzentsenko, Vladimiros and Zioulis, Nikolaos and Albanis, Georgios and Zarpalas, Dimitrios},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={3716--3726},
  year={2021}
}

Acknowledgements

This project has received funding from the European Union's Horizon 2020 innovation programme ATLANTIS under grant agreement No 951900.

Our code borrows from SEAN and deepfillv2.