DR-Pose

DRPose is developed on Ubuntu 18.04 OS, 3090 Ti GPU (CUDA 11.3). To reproduce our results:

1. Create conda environment

conda create -n DRPose python=3.6
conda activate DRPose

2. Install basic packages

cd DRPose
pip install -r requirements.txt

3. install PyTorch + CUDA

pip install torch==1.10.1+cu113 torchvision==0.11.2+cu113 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu113/torch_stable.html

4. By running the following shell script, the other needed packages can be installed/compiled.

bash installer.sh

5. Download weights and test dataset from anonymous link:

https://drive.google.com/drive/folders/1DQ0GDCEBgRXH-_yebx31ez5_Doatf9du?usp=sharing

Extract weights.zip under /DRPose/weights folder. The architecture should be:

|DRPose
 -|weights
 ---|registration
 ---|deformation
 ---|completion

Extract NOCS.zip under /DRPose/dataset/NOCS folder. The architecture should be:

|DRPose
 -|dataset
 ---|NOCS
 -----|CAMERA
 -----|gts
 -----|obj_models
 -----|Real
 -----|results

6. For evaluaiton on CAMERA25 dataset:

python evaluate.py --dataset=camera_val

7. For evaluation on REAL275 dataset:

python evaluate.py --dataset=real_test