from sldl.video import VideoSR from sldl.image import ImageSR import gradio as gr import tempfile import shutil import torch import ffmpeg import time from PIL import Image cc = 2 if torch.backends.mps.is_available(): device = 'mps' cc = 5 elif torch.cuda.is_available(): device = 'cuda' cc = 10 else: device = 'cpu' vbsrgan = VideoSR('BSRGAN').to(device) vresrgan = VideoSR('RealESRGAN').to(device) ibsrgan = ImageSR('BSRGAN').to(device) iresrgan = ImageSR('RealESRGAN').to(device) def upscale_video(input_video, output_video, progress, mname): modelname = mname.lower() model = vbsrgan if modelname == 'bsrgan (default)': # do nothing pass elif modelname == 'real esrgan': model = vresrgan model(input_video, output_video, progress.tqdm) def upscale_image(input_image, output_image, mname): modelname = mname.lower() model = ibsrgan if modelname == 'bsrgan (default)': # do nothing pass elif modelname == 'real esrgan': model = iresrgan shutil.copy(input_image, output_image) model(output_image) # Gradio interface def video_upscaling_interface(input_text, model_name, progress=gr.Progress()): if input_text: temp_dir = tempfile.mkdtemp() input_video_path = f"{temp_dir}/input_video" output_video_path = f"{temp_dir}/output_video.mp4" ffmpeg.input(input_text).output(input_video_path + '.mp4').run() # Upscale the video upscale_video(input_video_path + '.mp4', output_video_path, progress, model_name) return [output_video_path, output_video_path] else: return ["no_vid.mp4", "no_vid.mp4"] def image_upscaling_interface(input_text, model_name): if input_text: temp_dir = tempfile.mkdtemp() input_image_path = f"{temp_dir}/input_image.jpg" output_image_path = f"{temp_dir}/output_image.jpg" input_text.save(input_image_path) upscale_image(input_image_path, output_image_path, model_name) return [output_image_path, output_image_path] else: return ["no_image.jpg", "no_image.jpg"] css = "footer {display: none !important;} .gradio-container {min-height: 0px !important;}" with gr.Blocks(css=css) as demo: gr.Markdown(''' # Upscale ## A CVSYS Project ### NOTICE: This is running on a free Hugging Face Space, so it will be quite slow. Expect it to take _hours_ to upscale 5 minutes. Please be mindful and _DO NOT_ upscale videos longer than 15 seconds! Thank you! [Check out Upscale on GitHub!](https://github.com/cv-sys/upscale) ## Want Faster Inference? Duplicate this space for faster inference! We recommend using an A10G or A100. Please note that after you upload an image, it may take several minutes before the progress bar appears. This is because we first convert your video to ensure the correct format. ''') # with gr.Tab("Image"): # with gr.Row(): # with gr.Column(): # iinp = gr.Image(label="Upload Image", interactive=True, type="pil") # imod = gr.Dropdown( # ["BSRGAN (Default)", "Real ESRGAN"], # value="BSRGAN (Default)", # interactive=True, # label="Model" # ) # with gr.Column(): # iout = gr.Image(label="View Image", interactive=False, type="filepath") # ifile = gr.File(label="Download Image", interactive=False) # ibtn = gr.Button(value="Upscale Image") with gr.Tab("Video"): with gr.Row(): with gr.Column(): vinp = gr.Video(label="Upload Video", interactive=True) vmod = gr.Dropdown( ["BSRGAN (Default)", "Real ESRGAN"], value="BSRGAN (Default)", interactive=True, label="Model" ) with gr.Column(): vout = gr.Video(label="Watch Video", interactive=False) vfile = gr.File(label="Download Video", interactive=False) vbtn = gr.Button(value="Upscale Video") # ibtn.click(image_upscaling_interface, [iinp, imod], outputs=[iout, ifile]) vbtn.click(video_upscaling_interface, [vinp, vmod], outputs=[vout, vfile]) demo.queue(concurrency_count=cc) demo.launch()