【自己幫影片換主角】FakeApp 教學

個model每100下自動save架喎

今晚再睇下先

ffmpeg -f image2 -i out%d.png -vcodec libx264 -crf 15 -pix_fmt yuv420p output.mp4

請問最後一個步驟合返埋得6秒點搞

你原片幾長

我撚知你係咪本身得6秒

連第一步extract都唔得

DATA_A
RuntimeError: module compiled against API version 0xc but this version of numpy is 0xb
0%| | 0/1545 [00:00<?, ?it/s]
Traceback (most recent call last):
File "align_faces.py", line 145, in <module>
File "align_faces.py", line 136, in main
File "align_faces.py", line 116, in iter_face_alignments
File "face_alignment\api.py", line 172, in get_landmarks
File "face_alignment\api.py", line 160, in detect_faces
RuntimeError: Error while calling cudaOccupancyMaxPotentialBlockSize(&num_blocks,&num_threads,K) in file c:\arthur\dlib-master\dlib\dnn\cuda_utils.h:155. code: 8, reason: invalid device function
[2152] Failed to execute script align_faces

DATA_B
RuntimeError: module compiled against API version 0xc but this version of numpy is 0xb
0%| | 0/1 [00:00<?, ?it/s]libpng warning: iCCP: cHRM chunk does not match sRGB
libpng warning: iCCP: cHRM chunk does not match sRGB

Traceback (most recent call last):
File "align_faces.py", line 145, in <module>
File "align_faces.py", line 136, in main
File "align_faces.py", line 116, in iter_face_alignments
File "face_alignment\api.py", line 172, in get_landmarks
File "face_alignment\api.py", line 160, in detect_faces
RuntimeError: Unsupported image type, must be 8bit gray or RGB image.
[240] Failed to execute script align_faces

求救

data_a 有人頭cap唔到.....
搞到out%d.png 斷開

DATA_A 冇cuda
DATA_B 圖片唔係8bit gray or RGB

A 有裝cuda
B 圖片唔係8bit gray or RGB, 唔明, 我係png...

試左長片
暫時個人覺得最快捷係

1.
data A片 (當佢20分鐘片)
ffmpeg -i sceneA.mp4 -vf fps=3 "out%d.png"

2.
data B片 (當佢5分鐘)
ffmpeg -i sceneA.mp4 -vf fps=10 "out%d.png"

3.Extract
4.Train

5A. 再改名data_A做data_AO (純粹備用, 可以直接del)
5B. 開多個新data_A放返data_A原片入去
5C. 再 ffmpeg -i sceneA.mp4 -vf fps=30 "out%d.png"
5D. 再Extract Data A (要Gen個alignments.json檔)

6. Convert model+data A

點解我本身A同B片都係15秒cut出黎400幾張png, convert完都有400幾張png, 用ffmpeg合返段片得1秒幾?

檔名斷左次序

推!
幫幫手

http://sendvid.com/17qrhl4h
搵到Angelababy

得最尾個closeup似

令我諗起x man個妖后

有冇原片seed

請問原片同素材係咪唔洗一樣長度?
原片要10-15秒, 分左400幾張
素材係咪可以多過400幾張? 係咪可以準啲?

RuntimeError: module compiled against API ve
is 0xb
Traceback (most recent call last):
File "align_faces.py", line 7, in <module>
File "C:\FakeApp\alignment\torch\__init__.
from torch._C import *
ImportError: DLL load failed: The specified
[128] Failed to execute script align_faces

help

去系統SET返CUDA環境變數
https://dacota.tw/blog/post/deepfakes-fakeapp

裝C++ 2015

仲有無長d

裝完都仲係出呢個