bluehotel/CodeFormer

1"""2This file is used for deploying hugging face demo:3https://huggingface.co/spaces/sczhou/CodeFormer4"""5 6import sys7sys.path.append('CodeFormer')8import os9import cv210import torch11import torch.nn.functional as F12import gradio as gr13 14from torchvision.transforms.functional import normalize15 16from basicsr.utils import imwrite, img2tensor, tensor2img17from basicsr.utils.download_util import load_file_from_url18from facelib.utils.face_restoration_helper import FaceRestoreHelper19from facelib.utils.misc import is_gray20from basicsr.archs.rrdbnet_arch import RRDBNet21from basicsr.utils.realesrgan_utils import RealESRGANer22 23from basicsr.utils.registry import ARCH_REGISTRY24 25 26os.system("pip freeze")27 28pretrain_model_url = {29    'codeformer': 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth',30    'detection': 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/detection_Resnet50_Final.pth',31    'parsing': 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/parsing_parsenet.pth',32    'realesrgan': 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/RealESRGAN_x2plus.pth'33}34# download weights35if not os.path.exists('CodeFormer/weights/CodeFormer/codeformer.pth'):36    load_file_from_url(url=pretrain_model_url['codeformer'], model_dir='CodeFormer/weights/CodeFormer', progress=True, file_name=None)37if not os.path.exists('CodeFormer/weights/facelib/detection_Resnet50_Final.pth'):38    load_file_from_url(url=pretrain_model_url['detection'], model_dir='CodeFormer/weights/facelib', progress=True, file_name=None)39if not os.path.exists('CodeFormer/weights/facelib/parsing_parsenet.pth'):40    load_file_from_url(url=pretrain_model_url['parsing'], model_dir='CodeFormer/weights/facelib', progress=True, file_name=None)41if not os.path.exists('CodeFormer/weights/realesrgan/RealESRGAN_x2plus.pth'):42    load_file_from_url(url=pretrain_model_url['realesrgan'], model_dir='CodeFormer/weights/realesrgan', progress=True, file_name=None)43 44# download images45torch.hub.download_url_to_file(46    'https://replicate.com/api/models/sczhou/codeformer/files/fa3fe3d1-76b0-4ca8-ac0d-0a925cb0ff54/06.png',47    '01.png')48torch.hub.download_url_to_file(49    'https://replicate.com/api/models/sczhou/codeformer/files/a1daba8e-af14-4b00-86a4-69cec9619b53/04.jpg',50    '02.jpg')51torch.hub.download_url_to_file(52    'https://replicate.com/api/models/sczhou/codeformer/files/542d64f9-1712-4de7-85f7-3863009a7c3d/03.jpg',53    '03.jpg')54torch.hub.download_url_to_file(55    'https://replicate.com/api/models/sczhou/codeformer/files/a11098b0-a18a-4c02-a19a-9a7045d68426/010.jpg',56    '04.jpg')57torch.hub.download_url_to_file(58    'https://replicate.com/api/models/sczhou/codeformer/files/7cf19c2c-e0cf-4712-9af8-cf5bdbb8d0ee/012.jpg',59    '05.jpg')60torch.hub.download_url_to_file(61    'https://raw.githubusercontent.com/sczhou/CodeFormer/master/inputs/cropped_faces/0729.png',62    '06.png')63 64def imread(img_path):65    img = cv2.imread(img_path)66    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)67    return img68 69# set enhancer with RealESRGAN70def set_realesrgan():71    half = True if torch.cuda.is_available() else False72    model = RRDBNet(73        num_in_ch=3,74        num_out_ch=3,75        num_feat=64,76        num_block=23,77        num_grow_ch=32,78        scale=2,79    )80    upsampler = RealESRGANer(81        scale=2,82        model_path="CodeFormer/weights/realesrgan/RealESRGAN_x2plus.pth",83        model=model,84        tile=400,85        tile_pad=40,86        pre_pad=0,87        half=half,88    )89    return upsampler90 91upsampler = set_realesrgan()92device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')93codeformer_net = ARCH_REGISTRY.get("CodeFormer")(94    dim_embd=512,95    codebook_size=1024,96    n_head=8,97    n_layers=9,98    connect_list=["32", "64", "128", "256"],99).to(device)100ckpt_path = "CodeFormer/weights/CodeFormer/codeformer.pth"101checkpoint = torch.load(ckpt_path)["params_ema"]102codeformer_net.load_state_dict(checkpoint)103codeformer_net.eval()104 105os.makedirs('output', exist_ok=True)106 107def inference(image, face_align, background_enhance, face_upsample, upscale, codeformer_fidelity):108    """Run a single prediction on the model"""109    try: # global try110        # take the default setting for the demo111        only_center_face = False112        draw_box = False113        detection_model = "retinaface_resnet50"114 115        print('Inp:', image, background_enhance, face_upsample, upscale, codeformer_fidelity)116        face_align = face_align if face_align is not None else True117        background_enhance = background_enhance if background_enhance is not None else True118        face_upsample = face_upsample if face_upsample is not None else True119        upscale = upscale if (upscale is not None and upscale > 0) else 2120 121        has_aligned = not face_align122        upscale = 1 if has_aligned else upscale123 124        img = cv2.imread(str(image), cv2.IMREAD_COLOR)125        print('\timage size:', img.shape)126 127        upscale = int(upscale) # convert type to int128        if upscale > 4: # avoid memory exceeded due to too large upscale129            upscale = 4 130        if upscale > 2 and max(img.shape[:2])>1000: # avoid memory exceeded due to too large img resolution131            upscale = 2 132        if max(img.shape[:2]) > 1500: # avoid memory exceeded due to too large img resolution133            upscale = 1134            background_enhance = False135            face_upsample = False136 137        face_helper = FaceRestoreHelper(138            upscale,139            face_size=512,140            crop_ratio=(1, 1),141            det_model=detection_model,142            save_ext="png",143            use_parse=True,144            device=device,145        )146        bg_upsampler = upsampler if background_enhance else None147        face_upsampler = upsampler if face_upsample else None148 149        if has_aligned:150            # the input faces are already cropped and aligned151            img = cv2.resize(img, (512, 512), interpolation=cv2.INTER_LINEAR)152            face_helper.is_gray = is_gray(img, threshold=5)153            if face_helper.is_gray:154                print('\tgrayscale input: True')155            face_helper.cropped_faces = [img]156        else:157            face_helper.read_image(img)158            # get face landmarks for each face159            num_det_faces = face_helper.get_face_landmarks_5(160            only_center_face=only_center_face, resize=640, eye_dist_threshold=5161            )162            print(f'\tdetect {num_det_faces} faces')163            # align and warp each face164            face_helper.align_warp_face()165 166        # face restoration for each cropped face167        for idx, cropped_face in enumerate(face_helper.cropped_faces):168            # prepare data169            cropped_face_t = img2tensor(170                cropped_face / 255.0, bgr2rgb=True, float32=True171            )172            normalize(cropped_face_t, (0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True)173            cropped_face_t = cropped_face_t.unsqueeze(0).to(device)174 175            try:176                with torch.no_grad():177                    output = codeformer_net(178                        cropped_face_t, w=codeformer_fidelity, adain=True179                    )[0]180                    restored_face = tensor2img(output, rgb2bgr=True, min_max=(-1, 1))181                del output182                torch.cuda.empty_cache()183            except RuntimeError as error:184                print(f"Failed inference for CodeFormer: {error}")185                restored_face = tensor2img(186                    cropped_face_t, rgb2bgr=True, min_max=(-1, 1)187                )188 189            restored_face = restored_face.astype("uint8")190            face_helper.add_restored_face(restored_face)191 192        # paste_back193        if not has_aligned:194            # upsample the background195            if bg_upsampler is not None:196                # Now only support RealESRGAN for upsampling background197                bg_img = bg_upsampler.enhance(img, outscale=upscale)[0]198            else:199                bg_img = None200            face_helper.get_inverse_affine(None)201            # paste each restored face to the input image202            if face_upsample and face_upsampler is not None:203                restored_img = face_helper.paste_faces_to_input_image(204                    upsample_img=bg_img,205                    draw_box=draw_box,206                    face_upsampler=face_upsampler,207                )208            else:209                restored_img = face_helper.paste_faces_to_input_image(210                    upsample_img=bg_img, draw_box=draw_box211                )212        else:213            restored_img = restored_face214 215        # save restored img216        save_path = f'output/out.png'217        imwrite(restored_img, str(save_path))218 219        restored_img = cv2.cvtColor(restored_img, cv2.COLOR_BGR2RGB)220        return restored_img221    except Exception as error:222        print('Global exception', error)223        return None, None224 225 226title = "CodeFormer: Robust Face Restoration and Enhancement Network"227 228description = r"""<center><img src='https://user-images.githubusercontent.com/14334509/189166076-94bb2cac-4f4e-40fb-a69f-66709e3d98f5.png' alt='CodeFormer logo'></center>229<br>230<b>Official Gradio demo</b> for <a href='https://github.com/sczhou/CodeFormer' target='_blank'><b>Towards Robust Blind Face Restoration with Codebook Lookup Transformer (NeurIPS 2022)</b></a><br>231🔥 CodeFormer is a robust face restoration algorithm for old photos or AI-generated faces.<br>232🤗 Try CodeFormer for improved stable-diffusion generation!<br>233"""234 235article = r"""236If CodeFormer is helpful, please help to ⭐ the <a href='https://github.com/sczhou/CodeFormer' target='_blank'>Github Repo</a>. Thanks! 237[![GitHub Stars](https://img.shields.io/github/stars/sczhou/CodeFormer?style=social)](https://github.com/sczhou/CodeFormer)238 239---240 241📝 **Citation**242 243If our work is useful for your research, please consider citing:244```bibtex245@inproceedings{zhou2022codeformer,246    author = {Zhou, Shangchen and Chan, Kelvin C.K. and Li, Chongyi and Loy, Chen Change},247    title = {Towards Robust Blind Face Restoration with Codebook Lookup TransFormer},248    booktitle = {NeurIPS},249    year = {2022}250}251```252 253📋 **License**254 255This project is licensed under <a rel="license" href="https://github.com/sczhou/CodeFormer/blob/master/LICENSE">S-Lab License 1.0</a>. 256Redistribution and use for non-commercial purposes should follow this license.257 258📧 **Contact**259 260If you have any questions, please feel free to reach me out at <b>shangchenzhou@gmail.com</b>.261 262🤗 **Find Me:**263<style type="text/css">264td {265    padding-right: 0px !important;266}267</style>268 269<table>270<tr>271    <td><a href="https://github.com/sczhou"><img style="margin:-0.8em 0 2em 0" src="https://img.shields.io/github/followers/sczhou?style=social" alt="Github Follow"></a></td>272    <td><a href="https://twitter.com/ShangchenZhou"><img style="margin:-0.8em 0 2em 0" src="https://img.shields.io/twitter/follow/ShangchenZhou?label=%40ShangchenZhou&style=social" alt="Twitter Follow"></a></td>273</tr>274</table>275 276<center><img src='https://api.infinitescript.com/badgen/count?name=sczhou/CodeFormer&ltext=Visitors&color=6dc9aa' alt='visitors'></center>277"""278 279demo = gr.Interface(280    inference, [281        gr.Image(type="filepath", label="Input"),282        gr.Checkbox(value=True, label="Pre_Face_Align"),283        gr.Checkbox(value=True, label="Background_Enhance"),284        gr.Checkbox(value=True, label="Face_Upsample"),285        gr.Number(value=2, label="Rescaling_Factor (up to 4)"),286        gr.Slider(0, 1, value=0.5, step=0.01, label='Codeformer_Fidelity (0 for better quality, 1 for better identity)')287    ], [288        gr.Image(type="numpy", label="Output").style(height='auto')289    ],290    title=title,291    description=description,292    article=article,       293    examples=[294        ['01.png', True, True, True, 2, 0.7],295        ['02.jpg', True, True, True, 2, 0.7],296        ['03.jpg', True, True, True, 2, 0.7],297        ['04.jpg', True, True, True, 2, 0.1],298        ['05.jpg', True, True, True, 2, 0.1],299        ['06.png', False, True, True, 1, 0.5]300      ])301 302DEBUG = os.getenv('DEBUG') == '1'303demo.queue(api_open=False, concurrency_count=2, max_size=10)304demo.launch(debug=DEBUG)305# demo.launch(debug=DEBUG, share=True)