Closed wackxu closed 5 years ago
For best performance use face_swap_single2many or copy its pipeline. The following parameters are most significant:
seg_model = ../data/face_seg_fcn8s_300.caffemodel # try the 500x500 model
seg_deploy = ../data/face_seg_fcn8s_300_deploy.prototxt # try the 500x500 model
generic = 0 # use generic mode (disable shape regression)
source_max_res = 500 # sources max resolution (try other resolutions)
target_max_res = 500 # targets max resolution (try other resolutions)
@YuvalNirkin If we use our data to retrain the model, will there be a significant improvement in performance?
@YuvalNirkin Could you give me some suggestions?