antgroup / echomimic

EchoMimic: Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioning
https://antgroup.github.io/ai/echomimic/
Apache License 2.0
3.06k stars 356 forks source link

A bug was found here, some jpeg format found no face here #137

Open XeoOuYang opened 3 months ago

XeoOuYang commented 3 months ago

face musk prepare

face_img = cv2.imread(ref_image_path) face_mask = np.zeros((face_img.shape[0], face_img.shape[1])).astype('uint8')

det_bboxes, probs = face_detector.detect(face_img)

det_bboxes will be None, once we use Image.open the same jpeg file, all things go well.

EXAMPLE file is here, I don't know if it can be reproduce or not after upload and download procedure.

test_03_25_h264_yuv420p_f0_blend.zip

--ZIP instead--

oisilener1982 commented 3 months ago

is there a solution for this

agchaowanhui commented 2 months ago

det_bboxes, probs = face_detector.detect(cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB))