configs/audio2motion/model/crop_config.yaml

insightface_root: pretrain_weights/decode/v1/insightface
landmark_ckpt_path: pretrain_weights/decode/v1/first_stage/landmark.onnx
xpose_config_file_path: src/utils/UniPose_SwinT.py
device_id: 0                         # gpu device id
flag_force_cpu: False                # force cpu inference, WIP
det_thresh: 0.15                      # detection threshold
########## source image or video cropping option ##########
dsize: 512                           # crop size
scale: 2.3                           # scale factor
vx_ratio: 0                          # vx ratio
vy_ratio: -0.125                     # vy ratio +up, -down
max_face_num: 0                      # max face number, 0 mean no limit
flag_do_rot: True                    # whether to conduct the rotation when flag_do_crop is True
animal_face_type: animal_face_9      # animal_face_68 -> 68 landmark points, animal_face_9 -> 9 landmarks
########## driving video auto cropping option ##########
scale_crop_driving_video: 2.2        # 2.0 # scale factor for cropping driving video
vx_ratio_crop_driving_video: 0.0     # adjust x offset
vy_ratio_crop_driving_video: -0.1  # adjust y offset
direction: large-small               # direction of cropping
source_max_dim: 1280
source_division: 2