search

ai-forever / ghost

A new one shot face swap approach for image and video domains
Apache License 2.0
1.25k stars 264 forks source link

Incorrect keypoint array indexing in video_processing.py #101

Open dannewbee opened 2 months ago

dannewbee commented 2 months ago

Description

There is an issue in the crop_frames_and_get_transforms function in util/inference/video_processing.py. The keypoint array indexing is incorrect, which causes the keypoints to be appended to the wrong list. When target face count is set to be >1 (multiple faces), result video would be wrong.

Current Code

for q in range (len(target_embeds)):                
    kps_array[0].append([])

Suggested Fix

The indexing should be corrected to append the keypoints to the correct list:

for q in range (len(target_embeds)):                
    kps_array[q].append([])