All the related data for data preparation can be downloaded here. You could download them first and then follow the instructions below for data preparation.
First, the following dataset need to be downloaded and extracted to the folder EXPDATA/
The preprocessed data following DeepIM and PVNet can be downloaded from LM6d_converted and raw_data. After downloading, you should put the downloaded files into the folder EXPDATA/ (lying in the repository's root directory). To create occluded objects during training, we follow PVNet to randomly create occlusions. You could run the following scripts to transform the data format for our dataloader.
bash scripts/run_dataformatter.sh
The command above will automatically save the formatted data into EXPDATA/.
You also need to download the object models and put the extracted folder models into *./EXPDATA/LM6d_converted/.
Pascal VOC 2012 need to be downloaded to folder EXPDATA/. These images will be necessary for the random background generation for training.
The initial poses estimated by PoseCNN and PVNet can be downloaded from here. The initial pose folder also should be put into the folder EXPDATA/
Run the following script to generate the info files, which is put into the folder EXPDATA/data_info/
bash scripts/run_datainfo_generation.sh
After the the data preparation, the expected directory structure should be
./EXPDATA
|──LM6d_converted
| |──LM6d_refine
| |──LM6d_refine_syn
| └──models
|──LINEMOD
| └──fuse_formatted
|──lmo
|──VOCdevkit
|──raw_data
|──init_poses
└──data_info