search

youngLBW / HRN

[CVPR2023] A Hierarchical Representation Network for Accurate and Detailed Face Reconstruction from In-The-Wild Images.
https://younglbw.github.io/HRN-homepage/
Apache License 2.0
407 stars 37 forks source link

Cmd command takes 25 sec while modelscope takes 1 sec #33

Open nailcankaratelegram opened 1 year ago

nailcankaratelegram commented 1 year ago

Hi. Im new in python. Below command takes 25 sec to producing results. 100% 1/1 [00:26<00:00, 26.75s/it] CUDA_VISIBLE_DEVICES=0 python demo.py --input_type single_view --input_root ./assets/examples/single_view_image --output_root ./assets/examples/single_view_image_results

But with the modelscope you provided in googlecolab in homepage, it takes 1 sec to producing results.

face_reconstruction = pipeline(Tasks.face_reconstruction, model='damo/cv_resnet50_face-reconstruction', model_revision='v2.0.0-HRN')
result = face_reconstruction('first_frame.jpg')

Neither was the first run.

  1. Why does it take so long to run with the first method?
  2. First method also accepts multiple_view, while second method accepts only one image. Is there any way to give multiple images?
  3. First method creates both mid and high-frequency mesh while second method creates only mid-frequency mesh. Is there a high-frequency mesh generation method in the second method? Any code changes?
  4. My nose is crooked. HRN does not represent correctly high curved things like my nose from front view. I know the faces in the dataset it was trained on were shapely faces, but wouldn't it adequately represent people with facial paralysis?
  5. What I wrote in item 4 was for photographs taken from the front view. In side view If the nose is arched, the reprensentation does not fit properly. image

Thanks.