Odd GPEN Error - 'mask_sharp' referenced before assignment

[indianajson]

When running inference.py the following error occurs when Step 5 begins

  File "/inference.py", line 342, in <module>
    main()
  File "/inference.py", line 198, in main
    pred, _, _ = enhancer.process(img, img, face_enhance=True, possion_blending=False)
  File "/third_part/GPEN/gpen_face_enhancer.py", line 116, in process
    mask_sharp = cv2.GaussianBlur(mask_sharp, (0,0), sigmaX=1, sigmaY=1, borderType = cv2.BORDER_DEFAULT)
UnboundLocalError: local variable 'mask_sharp' referenced before assignment

What's likely the cause?

[kunncheng]

Sorry for the confusing error.
The reason seems to be that one frame in the video fails to detect a face.
The GPEN detection code is here.

[indianajson]

Thanks for getting back to me, I bypassed that bit of code and ran into more errors. Eventually, I got an error telling me "a face wasn't detected". I didn't realize that was causing all the issues, thank you for telling me!

In situations like this, what's the best course of action? A different source video?

[kunncheng]

First you should check if there are any frames in your input video that don't have faces. You should cut out these clips.

In some cases, even though the input video contains full faces, the pre-trained detector still fails to detect them.
In my experience, when the face resolution is too large or the face area proportion is too large, it may not be detected. This may be related to the face detector's training set.

[indianajson]

The source definitely has a face in every frame, but I think it may be face area proportion. I cropped the video quite a bit for my test and I'll bet that's why it's having trouble.

Thanks so much for your help and fast response. This project is super exciting and I'm anticipating being blown away by my own results... as soon as I get an actual render. 😅

[PanJinquan]

fix this bug:
mask_sharp = np.zeros((height, width), dtype=np.float32)

[blackjxxx]

fix this bug: mask_sharp = np.zeros((height, width), dtype=np.float32)

fix this bug: mask_sharp = np.zeros((height, width), dtype=np.float32)

Where to write this line of code

[Jeriousman]

@PanJinquan Can you specify where to insert that piece of code?

[caidewu]

@Jeriousman

[Someonetoldme584]

I cant see the temps ?

[Fillllllllllll]

@Jeriousman

Hi! If i'm using colab version - where should i put this?

[arielweinberger]

@Fillllllllllll
https://github.com/OpenTalker/video-retalking/blob/main/third_part/GPEN/gpen_face_enhancer.py