camera-style

Installation

Python 3.9.7 Cuda 11.5

1. Setup a conda environment:

conda create --name camera-style python=3.9 pip
conda activate camera-style

2. To install required modules and load pre-trained models launch the following commands (LIB_DIR is a global path to a directory that will contain all downloaded external modules and must be outside of the current repository):

./setup/setup_env.sh
./setup/load_models.sh
LIB_DIR=/path/to/dir ./setup/load_lib.sh
python ./setup/confifg_test.py

Note: the version of matplolib can be conflicting with sort repository. To avoid such issue, the script config/comment_lines.py launched in config/setup.sh will comment lines 22 and 23 of lib/sort/sort.py:

import matplotlib
matplotlib.use('TkAgg')

External modules

This project use the following external modules:

• DOPE
• LCR-Net V2
• mannequinchallenge
• camera-control
• SORT
• depth-distillation
• RAFT
• motion-detection

Dataset structure

data_unity/
├── flow_*flow-type*_pk/           # pickle external flows
├── flow_*flow-type*_pth/          # pth external flows
├── flow_*flow-type*_maxnorm_pth/  # max-norm preprocessed pth external flows
├── flow_*flow-type*_unit_pth/     # unit preprocessed pth external flows
├── flow_*flow-type*_videos/       # mp4 external flows
│
├── flow_unity_pth/                # pth unity flows
├── flow_unity_maxnorm_pth/        # max-norm preprocessed pth unity flows
├── flow_unity_unit_pth/           # unit preprocessed pth unity flows
├── flow_unity_videos/             # mp4 unity flows
│
├── raw_unity/
│   ├── archives/                  # archives files
│   ├── flow_frames/               # png unity flows
│   ├── raw_all/                   # unity depth, OF, raw frames
│   └── raw_frames/                # png unity raw frames
│
├── raw_videos/                    # mp4 raw frames
│
└── splits/                        # train/val/test splits
    ├── test.csv
    ├── train.csv
    └── val.csv

Run the following command to preprocess flows:

ROOT_DIR=/path/to/data_unity ./setup/preprocess_flows.sh