Major Patch for Face Restoration

[pawansharmaaaa]

The thing that's been nagging at me since this project kicked off was the subpar quality I was getting from GFPGAN and other face restoration tools. I've been consistently working with 512x512 faces, but after streamlining my process and running numerous tests, I've finally nailed down the ideal input size for these models.

Do let me know if you find this update helpful...

Full Changelog: v1.0...v1.2

---

This discussion was created from the release Major Patch for Face Restoration.

[love060701]

Great job! I will try it!
the colab Launch Lip_Wise in Cell has an error:

AttributeError Traceback (most recent call last)
in <cell line: 9>()
7
8 # Custom Modules
----> 9 import infer
10 import file_check
11

15 frames
/usr/local/lib/python3.10/dist-packages/numpy/testing/_private/utils.py in
55 IS_PYSTON = hasattr(sys, "pyston_version_info")
56 HAS_REFCOUNT = getattr(sys, 'getrefcount', None) is not None and not IS_PYSTON
---> 57 HAS_LAPACK64 = numpy.linalg._umath_linalg._ilp64
58
59 _OLD_PROMOTION = lambda: np._get_promotion_state() == 'legacy'

AttributeError: module 'numpy.linalg._umath_linalg' has no attribute '_ilp64'

[love060701]

It shakes a lot

[pawansharmaaaa]

I am looking for a solution, this is happening due to lack of temporal smoothing in the restoration models.