-
Notifications
You must be signed in to change notification settings - Fork 52
Description
I have an image containing multiple people and I would like to change the expression of an individual, then replace the original face in the image with the new facial expression.
I have a workflow that I feel with accomplish this BUT the image remains unchanged! I am extracting the individual's face (SEGS), changing the expression (PHM) and updating the facial expression in the original image (ReActor Fast Face Swap). The required face is updated marginally BUT not with the required facial expression!
Looking at the detailed console log, it appears that ReActor considers the input and source faces to be the same and as such it uses the Hashed Face Model! The console log entries are recorded below:
Prepare source...
0: 640x576 1 face, 68.6ms
Speed: 3.0ms preprocess, 68.6ms inference, 1.0ms postprocess per image at shape (1, 3, 640, 576)
w, h:(179, 209)
diff, min_diff, best_box:(66.44227600097656, 179, [133.18665, 5.4574842, 178.6979, 77.67295])
[ReActor] 10:39:05 - STATUS - Working: source face index [0], target face index [4]
[ReActor] 10:39:05 - INFO - Source Image MD5 Hash = cf54aab17792e9b3a5cacc615ade4987
[ReActor] 10:39:05 - INFO - Source Image the Same? False
[ReActor] 10:39:05 - STATUS - Analyzing Source Image...
set det-size: (640, 640)
warning: det_size is already set in detection model, ignore
[ReActor] 10:39:05 - INFO - Target Image MD5 Hash = d07382ec01911259510a993201bae644
[ReActor] 10:39:05 - INFO - Target Image the Same? True
[ReActor] 10:39:05 - STATUS - Using Hashed Target Face(s) Model...
[ReActor] 10:39:05 - STATUS - Swapping...
[ReActor] 10:39:06 - STATUS - --Done!--
[ReActor] 10:39:06 - STATUS - Restoring with codeformer-v0.1.0.pth | Face Size is set to 512
Am I correct in my assumption? How can I change this behaviour?
Thanks...
rhodian
