# setup for camera control
git clone https://github.com/ewrfcas/Uni3C.git
cd Uni3C

apt-get update && apt-get install tmux ffmpeg libsm6 libxext6 libglm-dev -y
conda create -n uni3c python=3.10
conda activate uni3c

pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
pip install carvekit --no-deps

# install pytorch3d
git clone https://github.com/facebookresearch/pytorch3d.git
cd pytorch3d && pip install -e .

# setup for alignment
cd third_party/GVHMR_realisdance
bash install.sh
cd ../GeoCalib && pip install -e .
cd ../..

python cam_render.py --reference_image "data/demo/flux-0.png" \
                     --output_path "outputs/demo_0" \
                     --traj_type "custom" \
                     --x_offset 0.25 \
                     --d_phi -50.0
                     
python cam_render.py --reference_image "data/demo/flux-1.png" \
                     --output_path "outputs/demo_1" \
                     --traj_type "free1"

CUDA_VISIBLE_DEVICES=0 python cam_control.py \
                      --reference_image "data/demo/flux-0.png" \
                      --render_path "outputs/demo_0" \
                      --output_path "outputs/demo_0/result.mp4" \
                      --prompt "The video features a cartoonish bear sitting at a school desk in a classroom setting."
                      
                      
CUDA_VISIBLE_DEVICES=0 python cam_control.py \
                      --reference_image "data/demo/flux-1.png" \
                      --render_path "outputs/demo_1" \
                      --output_path "outputs/demo_1/result.mp4" \
                      --prompt "The video features a living room."
                      
# for faster FSDP + Sequential parallel inference with 4 GPUs
CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=4 cam_control.py \
                      --reference_image "data/demo/flux-0.png" \
                      --render_path "outputs/demo_0" \
                      --output_path "outputs/demo_0/result_sp.mp4" \
                      --prompt "The video features a cartoonish bear sitting at a school desk in a classroom setting." \
                      --enable_sp \
                      --fsdp

# run for pose extraction (ensure you are in the path of 'third_party/GVHMR_realisdance')
cd third_party/GVHMR_realisdance
python extract_pose.py --video "../../data/demo_uni3c/video.mp4" \
                       --output_root "../../outputs/demo_uni3c"
                       
# back to the main path (Uni3C)                      
cd ../.. 

python alignment.py --input_image "data/demo_uni3c/reference.png" \
                    --input_path "outputs/demo_uni3c" \
                    --output_path "outputs/demo_uni3c_aligned" \
                    --traj_type "free1"

# Sngle GPU inference
python uni3c_inference.py --reference_image "data/demo_uni3c/reference.png" \
                          --render_path "outputs/demo_uni3c_aligned" \
                          --output_path "outputs/demo_uni3c_aligned/result.mp4" \
                          --prompt "A woman with curly hair, dressed in a dark hoodie and black pants, dances energetically on a sidewalk in front of a well-maintained building with large arched windows. The building, with a light-colored facade, is surrounded by lush green trees and bushes, creating a serene urban park setting. The woman's movements are rhythmic and dynamic, with her arms and legs in motion, adding a lively element to the tranquil environment. The sidewalk, made of gray rectangular paving stones, is bordered by a curb and runs parallel to the building. In the background, there are outdoor seating areas with black umbrellas and tables, and a few cars are parked near the building. The overall scene is vibrant and realistic, capturing a moment of urban leisure and nature."
                          
# Multiple GPU inference
CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=4 uni3c_inference.py \
                                      --reference_image "data/demo_uni3c/reference.png" \
                                      --render_path "outputs/demo_uni3c_aligned" \
                                      --output_path "outputs/demo_uni3c_aligned/result.mp4" \
                                      --prompt "A woman with curly hair, dressed in a dark hoodie and black pants, dances energetically on a sidewalk in front of a well-maintained building with large arched windows. The building, with a light-colored facade, is surrounded by lush green trees and bushes, creating a serene urban park setting. The woman's movements are rhythmic and dynamic, with her arms and legs in motion, adding a lively element to the tranquil environment. The sidewalk, made of gray rectangular paving stones, is bordered by a curb and runs parallel to the building. In the background, there are outdoor seating areas with black umbrellas and tables, and a few cars are parked near the building. The overall scene is vibrant and realistic, capturing a moment of urban leisure and nature." \
                                      --enable_sp

@article{cao2025uni3c,
        title={Uni3C: Unifying Precisely 3D-Enhanced Camera and Human Motion Controls for Video Generation},
        author={Cao, Chenjie and Zhou, Jingkai and Li, shikai and Liang, Jingyun and Yu, Chaohui and Wang, Fan and Xue, Xiangyang and Fu, Yanwei},
        journal={arXiv preprint arXiv:2504.14899},
        year={2025}}