Object_Reconstruction

1) Data Download

Download Pretrained Weights for Segmentation Network

XMem-s012.pth

place them under ./XMem/saves/

Download Pretrained Weights for LoFTR

outdoor_ds.ckpt

place them under ./BundleSDF/BundleTrack/LoFTR/weights/

2) Docker Setup

To set up the Docker environment, run the following command:

docker build --network host -t nvcr.io/nvidian/bundlesdf .

3) Methodology

Capture Data from RealSense Depth Camera

To capture data from a RealSense depth camera, run the rec_con_mask.py script.

1)Install librealsense SDK
2)Install dependencies for Xmem and Realsense
conda environment (optional)

conda create -n obj_recon python=3.10

pip install -r requirements.txt

Prepare Input Directory

Run

conda info --base

to get conda base directory. Copy the path and paste it in capture.sh. Then, run the following.

sudo bash capture.sh

Note : (Don't run in Docker Container)

Press Enter to start recording RGB and depth frames. Then, create a boundary by selecting points on the window to create a mask for the first image.

Steps to Run

1. Run the container:

   bash run_container.sh
   

2. Build the project:

   bash build.sh
   

3. Run the project:

   bash run.sh