onebitss/Real-ESRGAN

1from PIL import Image2import cv2 as cv3import torch4from RealESRGAN import RealESRGAN5import tempfile6import numpy as np7import tqdm8import ffmpeg9import spaces10 11 12device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')13 14@spaces.GPU(duration=60)15def infer_image(img: Image.Image, size_modifier: int ) -> Image.Image:16    if img is None:17        raise Exception("Image not uploaded")18    19    width, height = img.size20    21    if width >= 5000 or height >= 5000:22        raise Exception("The image is too large.")23 24    model = RealESRGAN(device, scale=size_modifier)25    model.load_weights(f'weights/RealESRGAN_x{size_modifier}.pth', download=False)26 27    result = model.predict(img.convert('RGB'))28    print(f"Image size ({device}): {size_modifier} ... OK")29    return result30 31@spaces.GPU(duration=300)32def infer_video(video_filepath: str, size_modifier: int) -> str:33    model = RealESRGAN(device, scale=size_modifier)34    model.load_weights(f'weights/RealESRGAN_x{size_modifier}.pth', download=False)35 36    cap = cv.VideoCapture(video_filepath)37    38    tmpfile = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)39    vid_output = tmpfile.name40    tmpfile.close()41 42    # Check if the input video has an audio stream43    probe = ffmpeg.probe(video_filepath)44    has_audio = any(stream['codec_type'] == 'audio' for stream in probe['streams'])45 46    if has_audio:47        # Extract audio from the input video48        audio_file = video_filepath.replace(".mp4", ".wav")49        ffmpeg.input(video_filepath).output(audio_file, format='wav', ac=1).run(overwrite_output=True)50 51    vid_writer = cv.VideoWriter(52        vid_output,53        fourcc=cv.VideoWriter.fourcc(*'mp4v'),54        fps=cap.get(cv.CAP_PROP_FPS),55        frameSize=(int(cap.get(cv.CAP_PROP_FRAME_WIDTH)) * size_modifier, int(cap.get(cv.CAP_PROP_FRAME_HEIGHT)) * size_modifier)56    )57 58    n_frames = int(cap.get(cv.CAP_PROP_FRAME_COUNT))59 60    for _ in tqdm.tqdm(range(n_frames)):61        ret, frame = cap.read()62        if not ret:63            break64 65        frame = cv.cvtColor(frame, cv.COLOR_BGR2RGB)66        frame = Image.fromarray(frame)67 68        upscaled_frame = model.predict(frame.convert('RGB'))69        70        upscaled_frame = np.array(upscaled_frame)71        upscaled_frame = cv.cvtColor(upscaled_frame, cv.COLOR_RGB2BGR)72 73        vid_writer.write(upscaled_frame)74 75    vid_writer.release()76 77    if has_audio:78        # Re-encode the video with the modified audio79        ffmpeg.input(vid_output).output(video_filepath.replace(".mp4", "_upscaled.mp4"), vcodec='libx264', acodec='aac', audio_bitrate='320k').run(overwrite_output=True)80 81        # Replace the original audio with the upscaled audio82        ffmpeg.input(audio_file).output(video_filepath.replace(".mp4", "_upscaled.mp4"), acodec='aac', audio_bitrate='320k').run(overwrite_output=True)83 84    print(f"Video file : {video_filepath}")85 86    return vid_output.replace(".mp4", "_upscaled.mp4") if has_audio else vid_output