def handle_image_request(image: types.Image):
if image.content is None and image.source is None:
raise HTTPException(
status_code=400,
detail="Either image as bytes or image source with uri needs to be sent!",
)
try:
init = time.time()
if image.content is not None:
image_bytes = image.content
image = io.bytes2image(base64.b64decode(image_bytes))
else:
image_bytes = io.read_fn(image.source.imageUri)
image = io.bytes2image(image_bytes)
image_bytes = base64.b64encode(image_bytes)
print(f"Elapsed reading image: {time.time()-init}")
except Exception as e:
raise HTTPException(
status_code=500,
detail="Image {} could not be read. Error: {}".format(
image.source.imageUri, e
),
)
return image, image_bytes
def run_inference(
request: types.Request,
credentials: HTTPAuthorizationCredentials = security.http_credentials,
):
print("************ Getting avatar image...")
avatar, _ = handle_image_request(request.avatar)
avatar = cv2.resize(avatar, model_input_size)
if avatar.ndim == 2:
avatar = np.tile(avatar[..., None], [1, 1, 3])
print("************ Done!")
print("************* Setting avatar image to model ...")
model.set_source_image(avatar)
print("************ Done!")
print("************* Getting video frames ...")
video_bytes = base64.b64decode(request.video.content)
video_frames = list(io.bytes2video(video_bytes, fps=request.fps))
print("************ Done!")
# video_frames = video_frames[:5]
video_name = uuid.uuid4().hex
io.write_fn(f"app/static/{video_name}_orig.webm", video_bytes)
video_path = f"app/static/{video_name}.mp4"
print("************* Getting audio Object ...")
audio = io.get_audio_obj(video_bytes)
print("************ Done!")
bbox = model.get_face_bbox(video_frames[0])
print("************* Getting transform video ...")
output_frames = model_funs.generate_video(
model,
video_frames,
merge=request.merge,
axis=request.axis,
verbose=True,
model_input_size=model_input_size,
horizontal_flip=request.flip,
relative=request.transferFace,
# relative=True,
crop_bbox=bbox,
# debug=False,
)
model.reset_frames()
print("************ Done!")
print("************* Getting video in moviepy ...")
def gen_video(t):
return output_frames[min(int(t * request.fps), len(output_frames) - 1)]
video = VideoClip(gen_video, duration=len(output_frames) / request.fps,)
print("************ Done!")
# Calculate min duration between audio and video (for some odd reason may differ)
final_duration = min(video.duration, audio.duration)
video = video.set_duration(final_duration)
audio = audio.set_duration(final_duration)
print("************* Setting audio to video ...")
video = video.set_audio(audio.set_duration(video.duration))
print("************ Done!")
print("************* Saving and decoding video ...")
video.write_videofile(video_path, fps=request.fps)
# output_frames = video_frames
# io.write_video(video_path, output_frames)
video_bytes = io.read_fn(video_path)
result = base64.b64encode(video_bytes).decode()
print("************ Done!")
return Response(video=types.Video(content=result))
if os.path.exists(test_image_path):
img = read_image(test_image_path)
img = cv2.resize(img, fx=0.5, fy=0.5, dsize=None)
image = image2bytes(img)
image = base64.b64encode(image).decode()
else:
image = None
# with tempfile.TemporaryFile(suffix=".mp4") as fp:
fp = "temp.mp4"
fp = "test.webm"
# video_frames = get_frames_from_camera(0)
# write_video(fp, video_frames)
video_bytes = read_fn(fp)
video = base64.b64encode(video_bytes).decode()
data = {
"avatar": {
"content": image,
"source": {
"imageUri": test_image_path
}
},
"video": {
"content": video
},
"merge": True,
"fps": 30,
"transferFace": True,
get_audio_obj,
)
import app.model_funs as model_funs
config_path = "fomm/config/vox-adv-256.yaml"
checkpoint_path = "vox-adv-cpk.pth.tar"
model = PredictorLocal(config_path, checkpoint_path, adapt_movement_scale=True)
IMG_SIZE = 256
size = (IMG_SIZE, IMG_SIZE)
avatar = cv2.imread("avatars/jobs.jpg")[..., ::-1]
avatar = cv2.imread("avatars/mona.jpg")[..., ::-1]
avatar = cv2.resize(avatar, size)
model.set_source_image(avatar)
video_path = "test.webm"
video_frames = list(bytes2video(read_fn(video_path)))
bbox = model.get_face_bbox(video_frames[0])
output_frames = model_funs.generate_video(
model,
video_frames,
merge=False,
verbose=True,
horizontal_flip=True,
relative=True,
crop_bbox=bbox,
debug=True,
)
get_audio_obj,
)
from moviepy.editor import *
import tempfile
video_path = "download.webm"
# video_path = "sisa.webm"
# audio = AudioFileClip(video_path)
# video = VideoFileClip(video_path, fps_source="fps")
# print(video)
# print(len([frame for frame in video.iter_frames()]))
# ffmpeg -i current.webm -c copy -fflags +genpts new.webm
video_bytes = read_fn(video_path)
print(get_audio_obj(video_bytes))
exit(0)
with tempfile.TemporaryDirectory() as temp_dir:
print(temp_dir)
tmp_video = temp_dir + "/sisa.webm"
tmp_video2 = temp_dir + "/sisa2.webm"
with open(tmp_video, "wb") as f:
f.write(video_bytes)
input = ffmpeg.input(tmp_video)
out = ffmpeg.output(input,
tmp_video2,
vcodec="copy",
acodec="copy",
