Awesome

iMapper: Interaction-guided Scene Mapping from Monocular Videos (link)

i3DB

Download videos and images

sh get_data.sh

Open a ground truth scenelet

Setup

conda create --name iMapper python=3 numpy -y

Usage

conda activate iMapper
export PYTHONPATH=$(pwd); python3 ./example.py

Run iMapper on a video

### Requirements

• CUDA capable GPU
• docker
• nvidia-docker

### How to run:

• Have a folder containing
  • a video (e.g., video.mp4) and
  • camera intrinsics in intrinsics.json e.g., [[1920.0, 0.0, 960.0], [0.0, 1920.0, 540.0], [0.0, 0.0, 1.0]]
• Replace the variables PATH_TO_FOLDER_CONTAINING_VIDEO and VIDEO below.
• Adjust GPU ID if needed.

git clone https://github.com/amonszpart/iMapper.git
cd iMapper
docker build -t imapper imapper/docker
nvidia-docker run -it --name iMapper \
 -v ${PATH_TO_FOLDER_CONTAINING_VIDEO}:/opt/iMapper/i3DB/MyScene:rw \
 imapper \
 bash -c "CUDA_VISIBLE_DEVICES=0 \
          python3 run_video.py /opt/iMapper/i3DB/MyScene/${VIDEO} --gpu-id 0"