I get the following error: RuntimeError: CUDA out of memory. Tried to allocate 88.00 MiB (GPU 0; 5.80 GiB total capacity; 4.14 GiB already allocated; 154.56 MiB free; 4.24 GiB reserved in total by PyTorch)

Is there a way to allocate more memory? I do not get why 4.14Gb are already allocated.

Could someone insert an installation and use tutorial for Windows 10? I have installed Python, Anaconda, PyThorc and Nvidia Cuda, but I didn't understand how to install and use this artificial intelligence software for restoring old photos (sorry if my English is bad but I'm Italian and I'm using Google Translate)

follow the jouranl edition of the paper, I wrote a pytorch Pseudocode training code

    # build 2 vae network, 3 discriminators but NO transfer network for now
    # and their optimizer
    vae1, xr_recon_d, z_xr_d, \
        vae2, y_recon_d, \
        optimizer_vae1, optimizer_d1, \
        optimizer_vae2, optimizer_d2 = build_model(opt)
    start_iter = 0
    if opt.load_checkpoint_iter>0:
        checkpoint_path = checkpoint_root + f'/global_checkpoint_{opt.load_checkpoint_iter}.pth'
        if not Path(checkpoint_path).exists():
            print(f"ERROR! checkpoint_path {checkpoint_path} is None")
            exit(-1)
        state_dict = torch.load(checkpoint_path)
        start_iter = state_dict['iter']
        assert state_dict['batch_size'] == opt.batch_size, f"ERROR - batch size changed! load: {state_dict['batch_size']}, but now {opt.batch_size}"
        vae1.load_state_dict(state_dict['vae1'])
        xr_recon_d.load_state_dict(state_dict['xr_recon_d'])
        z_xr_d.load_state_dict(state_dict['z_xr_d'])
        vae2.load_state_dict(state_dict['vae2'])
        y_recon_d.load_state_dict(state_dict['y_recon_d'])
        optimizer_vae1.load_state_dict(state_dict['optimizer_vae1'])
        optimizer_d1.load_state_dict(state_dict['optimizer_d1'])
        optimizer_vae2.load_state_dict(state_dict['optimizer_vae2']) 
        optimizer_d2.load_state_dict(state_dict['optimizer_d2']) 
        print("checkpoint load successfully!")
    # create dataloader
    dataLoaderR, dataLoaderXY = get_dataloader(opt)
    dataLoaderXY_iter = iter(dataLoaderXY)
    dataLoaderR_iter = iter(dataLoaderR)
    start = time.perf_counter()
    print("train start!")
    for ii in range(opt.total_iter - start_iter):
        current_iter = ii + start_iter
        try:
            x, y, path_y = dataLoaderXY_iter.next()
        except:
            dataLoaderXY_iter = iter(dataLoaderXY)
            x, y, path_y = dataLoaderXY_iter.next()
        try:
            r, path_r = dataLoaderR_iter.next()
        except:
            dataLoaderR_iter = iter(dataLoaderR)
            r, path_r = dataLoaderR_iter.next()
        ### following the practice in U-GAT-IT:
        ### train D and G iteratively, but not training D multiple times than training G
        r = r.to(opt.device)
        x = x.to(opt.device)
        y = y.to(opt.device)
        if opt.debug and current_iter%500==0:
            torchvision.utils.save_image(y, 'train_vae_y.png', normalize=True)
            torchvision.utils.save_image(x, 'train_vae_x.png', normalize=True)
            torchvision.utils.save_image(r, 'train_vae_r.png', normalize=True)

        ### vae1 train d
        # save gpu memory since no need calc grad for net G when train net D
        with torch.no_grad():
            z_x, mean_x, var_x, recon_x = vae1(x)
            z_r, mean_r, var_r, recon_r = vae1(r)
            batch_requires_grad(z_x, mean_x, var_x, recon_x,
                                z_r, mean_r, var_r, recon_r)
        loss_1 = 0
        adv_loss_d_x = lsgan_d(xr_recon_d(x), xr_recon_d(recon_x))
        adv_loss_d_r = lsgan_d(xr_recon_d(r), xr_recon_d(recon_r))
        # z_x is real and z_r is fake here because let z_r close to z_x
        adv_loss_d_xr = lsgan_d(z_xr_d(z_x), z_xr_d(z_r))
        loss_1_d = adv_loss_d_x + adv_loss_d_r + adv_loss_d_xr
        loss_1_d.backward()
        optimizer_d1.step()
        optimizer_d1.zero_grad()
        ### vae1 train g
        # since we need update weights of G, the result should be re-calculate with grad
        z_x, mean_x, var_x, recon_x = vae1(x)
        z_r, mean_r, var_r, recon_r = vae1(r)
        adv_loss_g_x = lsgan_g(xr_recon_d(recon_x))
        adv_loss_g_r = lsgan_g(xr_recon_d(recon_r))
        # z_x is real and z_r is fake here because let z_r close to z_x
        adv_loss_g_xr = lsgan_g(z_xr_d(z_r))
        KLDloss_1_x = -0.5 * torch.sum(1 + var_x - mean_x.pow(2) - var_x.exp())  # KLD
        L1loss_1_x  = opt.weight_alpha * F.l1_loss(x, recon_x)
        KLDloss_1_r = -0.5 * torch.sum(1 + var_r - mean_r.pow(2) - var_r.exp())  # KLD
        L1loss_1_r  = opt.weight_alpha * F.l1_loss(r, recon_r)
        loss_1_g = adv_loss_g_x + KLDloss_1_x + L1loss_1_x \
                 + adv_loss_g_r + KLDloss_1_r + L1loss_1_r \
                 + adv_loss_g_xr
        loss_1_g.backward()
        optimizer_vae1.step()
        optimizer_vae1.zero_grad()

        ### vae2 train d
        # save gpu memory since no need calc grad for net G when train net D
        with torch.no_grad():
            z_y, mean_y, var_y, recon_y = vae2(y)
            batch_requires_grad(z_y, mean_y, var_y, recon_y)
        adv_loss_d_y = lsgan_d(y_recon_d(y), y_recon_d(recon_y))
        loss_2_d = adv_loss_d_y
        loss_2_d.backward()
        optimizer_d2.step()
        optimizer_d2.zero_grad()
        ### vae2 train g
        # since we need update weights of G, the result should be re-calculate with grad
        z_y, mean_y, var_y, recon_y = vae2(y)
        adv_loss_g_y = lsgan_g(y_recon_d(recon_y))
        KLDloss_1_y = -0.5 * torch.sum(1 + var_y - mean_y.pow(2) - var_y.exp())  # KLD
        L1loss_1_y  = opt.weight_alpha * F.l1_loss(y, recon_y)
        loss_2_g = adv_loss_g_y + KLDloss_1_y + L1loss_1_y
        loss_2_g.backward()
        optimizer_vae2.step()
        optimizer_vae2.zero_grad()
        # debug
        if opt.debug and current_iter%500==0:
            # [print(k, 'channel 0:\n', v[0][0]) for k,v in list(model.named_parameters()) if k in ["netG_A.encoder.13.conv_block.5.weight", "netG_A.decoder.4.conv_block.5.weight"]]
            torchvision.utils.save_image(recon_x, 'train_vae_recon_x.png', normalize=True)
            torchvision.utils.save_image(recon_r, 'train_vae_recon_r.png', normalize=True)
            torchvision.utils.save_image(recon_y, 'train_vae_recon_y.png', normalize=True)
       
        if current_iter%500==0:
            print(f"""STEP {current_iter:06d} {time.perf_counter() - start:.1f} s
            loss_1_d = adv_loss_d_x + adv_loss_d_r + adv_loss_d_xr
            {loss_1_d:.3f} = {adv_loss_d_x:.3f} + {adv_loss_d_r:.3f} + {adv_loss_d_xr:.3f}
            loss_1_g = adv_loss_g_x + KLDloss_1_x + L1loss_1_x \
                 + adv_loss_g_r + KLDloss_1_r + L1loss_1_r \
                 + adv_loss_g_xr
            {loss_1_g:.3f} = {adv_loss_g_x:.3f} + {KLDloss_1_x:.3f} + {L1loss_1_x:.3f} \
                 + {adv_loss_g_r:.3f} + {KLDloss_1_r:.3f} + {L1loss_1_r:.3f} \
                 + {adv_loss_g_xr:.3f}
            """)
        if (current_iter+1)%2000==0:
            # finish the current_iter-th step, e.g. finish iter0, save as 1, resume train from iter 1
            state = {
                'iter': current_iter,
                'batch_size': opt.batch_size,
                #
                'vae1': vae1.state_dict(),
                'xr_recon_d': xr_recon_d.state_dict(),
                'z_xr_d': z_xr_d.state_dict(),
                #
                'vae2': vae2.state_dict(),
                'y_recon_d': y_recon_d.state_dict(),
                #
                'optimizer_vae1': optimizer_vae1.state_dict(),
                'optimizer_d1': optimizer_d1.state_dict(),
                'optimizer_vae2': optimizer_vae2.state_dict(),
                'optimizer_d2': optimizer_d2.state_dict(),
                }
            torch.save(state, checkpoint_root + f'/global_checkpoint_{current_iter}.pth')
    print("global", time.perf_counter() - start, ' s')

where the lsgan_d and lsgan_g is defined as following:

import torch
import torch.nn.functional as F
### lsgan: a=0, b=c=1
def lsgan_d(d_logit_real, d_logit_fake):
    return F.mse_loss(d_logit_real, torch.ones_like(d_logit_real)) + d_logit_fake.pow(2).mean()

def lsgan_g(d_logit_fake):
    return F.mse_loss(d_logit_fake, torch.ones_like(d_logit_fake))

Is it possible to run this on a (recent) Mac, which does not support CUDA? I would have guessed setting --GPU 0 would not attempt to call CUDA, but it fails.

File "/Users/../Desktop/bopbtl/venv/lib/python3.7/site-packages/torch/cuda/__init__.py", line 61, in _check_driver
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

Make a interactive frontend for the project Bringing Old Photos Back , where the user can upload photo or click it by camera and then can download the result image and share it.

Fixes: #125 Added script to apply filters on output image. Currently added script to make output image black and white. Will adding more scripts to more beautify images according to choice. This a feature which can help edit old images to look cool in this time.

Sample Video:

https://user-images.githubusercontent.com/65889104/112759421-875fa180-9010-11eb-87bd-66cd79222127.mp4

Running Stage 1: Overall restoration Now you are processing 5d6bfe49ge1b23ec32641&690.jpg Skip 5d6bfe49ge1b23ec32641&690.jpg due to an error: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR You can try to repro this exception using the following code snippet. If that doesn't trigger the error, please include your original repro script when reporting this issue.

import torch torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.benchmark = True torch.backends.cudnn.deterministic = False torch.backends.cudnn.allow_tf32 = True data = torch.randn([1, 3, 470, 446], dtype=torch.float, device='cuda', requires_grad=True) net = torch.nn.Conv2d(3, 64, kernel_size=[7, 7], padding=[0, 0], stride=[1, 1], dilation=[1, 1], groups=1) net = net.cuda().float() out = net(data) out.backward(torch.randn_like(out)) torch.cuda.synchronize()

ConvolutionParams data_type = CUDNN_DATA_FLOAT padding = [0, 0, 0] stride = [1, 1, 0] dilation = [1, 1, 0] groups = 1 deterministic = false allow_tf32 = true input: TensorDescriptor 0x54bd8730 type = CUDNN_DATA_FLOAT nbDims = 4 dimA = 1, 3, 470, 446, strideA = 628860, 209620, 446, 1, output: TensorDescriptor 0x54bda2f0 type = CUDNN_DATA_FLOAT nbDims = 4 dimA = 1, 64, 464, 440, strideA = 13066240, 204160, 440, 1, weight: FilterDescriptor 0x3b058e0 type = CUDNN_DATA_FLOAT tensor_format = CUDNN_TENSOR_NCHW nbDims = 4 dimA = 64, 3, 7, 7, Pointer addresses: input: 0x60b80c400 output: 0x60be60000 weight: 0x601660000

Skipping non-file final_output Skipping non-file stage_1_restore_output Skipping non-file stage_2_detection_output Skipping non-file stage_3_face_output Now you are processing timg.jpg Skip timg.jpg due to an error: CUDA error: an illegal memory access was encountered Finish Stage 1 ...

Is this because of insufficient video memory?

Need to manually create the restored_image dir otherwise the following failure occurs:

$ python run.py --input_folder ~/Documents/revive-old-photos/input --output ~/Documents/revive-old-photos/output --GPU 0
Running Stage 1: Overall restoration
Traceback (most recent call last):
  File "test.py", line 6, in <module>
    from torch.autograd import Variable
ImportError: No module named torch.autograd
Traceback (most recent call last):
  File "run.py", line 79, in <module>
    for x in os.listdir(stage_1_results):
FileNotFoundError: [Errno 2] No such file or directory: '~/Documents/revive-old-photos/output/stage_1_restore_output/restored_image'

I reimplement the training code, however, it is hard to increase SSIM to 0.69 when training VAE1 and VAE2 unless I change the L1 factor 'alpha'. I would like to know that how long did it take you to train VAE1 and VAE2? Would you tell me the rough time about releasing training code?

With some image throws this exception: Skip a.png due to an error: CUDA out of memory. Tried to allocate 3.10 GiB (GPU 0; 8.00 GiB total capacity; 3.83 GiB already allocated; 2.17 GiB free; 4.40 GiB reserved in total by PyTorch).

Thanks in advance

when i trying to install required libraries i got the following error:

Running setup.py install for torch ... error ERROR: Command errored out with exit status 1: command: 'c:\users\mahmed.ssmain\appdata\local\programs\python\python38\python.exe' -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\Users\mahmed .SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch\setup.py'"'"'; file='"'"'C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch\ setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(file);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, file, '"'"'exec '"'"'))' install --record 'C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-record-gm8ndtxb\install-record.txt' --single-version-externally-managed --compile --install-h eaders 'C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\Include\torch' cwd: C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch
Complete output (23 lines): running install running build_deps Traceback (most recent call last): File "", line 1, in File "C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch\setup.py", line 225, in setup(name="torch", version="0.1.2.post2", File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\site-packages\setuptools_init_.py", line 145, in setup return distutils.core.setup(**attrs) File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\distutils\core.py", line 148, in setup dist.run_commands() File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\distutils\dist.py", line 966, in run_commands self.run_command(cmd) File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\distutils\dist.py", line 985, in run_command cmd_obj.run() File "C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch\setup.py", line 99, in run self.run_command('build_deps') File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\distutils\cmd.py", line 313, in run_command self.distribution.run_command(command) File "C:\Users\mahmed.SSMAIN\AppData\Local\Programs\Python\Python38\lib\distutils\dist.py", line 985, in run_command cmd_obj.run() File "C:\Users\mahmed.SSMAIN\AppData\Local\Temp\pip-install-5a62zurd\torch\setup.py", line 51, in run from tools.nnwrap import generate_wrappers as generate_nn_wrappers ModuleNotFoundError: No module named 'tools.nnwrap'

--2022-12-28 03:36:54-- https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Face_Enhancement/checkpoints.zip Resolving facevc.blob.core.windows.net (facevc.blob.core.windows.net)... 20.60.228.1 Connecting to facevc.blob.core.windows.net (facevc.blob.core.windows.net)|20.60.228.1|:443... connected. HTTP request sent, awaiting response... 404 The specified resource does not exist. 2022-12-28 03:36:54 ERROR 404: The specified resource does not exist..

unzip: cannot find or open checkpoints.zip, checkpoints.zip.zip or checkpoints.zip.ZIP. /content/photo_restoration /content/photo_restoration/Global --2022-12-28 03:36:54-- https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Global/checkpoints.zip Resolving facevc.blob.core.windows.net (facevc.blob.core.windows.net)... 20.60.228.1 Connecting to facevc.blob.core.windows.net (facevc.blob.core.windows.net)|20.60.228.1|:443... connected. HTTP request sent, awaiting response... 404 The specified resource does not exist. 2022-12-28 03:36:55 ERROR 404: The specified resource does not exist..

Please help in this regards

┌──(root㉿kali)-[/home/kirisawa/Bringing-Old-Photos-Back-to-Life] └─# python3 run.py
Running Stage 1: Overall restoration Traceback (most recent call last): File "/home/kirisawa/Bringing-Old-Photos-Back-to-Life/Global/test.py", line 6, in from torch.autograd import Variable ModuleNotFoundError: No module named 'torch' Traceback (most recent call last): File "/home/kirisawa/Bringing-Old-Photos-Back-to-Life/run.py", line 102, in for x in os.listdir(stage_1_results): FileNotFoundError: [Errno 2] No such file or directory: '/home/kirisawa/Bringing-Old-Photos-Back-to-Life/output/stage_1_restore_output/restored_image'

I've run the requirements .txt but I'm still having problems.😭😭

Hello, author. First of all, thank you very much for the code you posted. I have a question for you, that is, I can't find the code that adds scratches. Can you tell me where it is?

# pull the syncBN repo
%cd photo_restoration/Face_Enhancement/models/networks
!git clone https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
!cp -rf Synchronized-BatchNorm-PyTorch/sync_batchnorm .
%cd ../../../

%cd Global/detection_models
!git clone https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
!cp -rf Synchronized-BatchNorm-PyTorch/sync_batchnorm .
%cd ../../

# download the landmark detection model
%cd Face_Detection/
!wget http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2
!bzip2 -d shape_predictor_68_face_landmarks.dat.bz2
%cd ../

# download the pretrained model
%cd Face_Enhancement/
!wget https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Face_Enhancement/checkpoints.zip
!unzip checkpoints.zip
%cd ../

%cd Global/
!wget https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Global/checkpoints.zip
!unzip checkpoints.zip
%cd ../

/content/photo_restoration
/content/photo_restoration/Face_Enhancement
--2022-11-09 21:01:48--  https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Face_Enhancement/checkpoints.zip
Resolving facevc.blob.core.windows.net (facevc.blob.core.windows.net)... 20.60.228.1
Connecting to facevc.blob.core.windows.net (facevc.blob.core.windows.net)|20.60.228.1|:443... connected.
HTTP request sent, awaiting response... 404 The specified resource does not exist.
2022-11-09 21:01:49 ERROR 404: The specified resource does not exist..

unzip:  cannot find or open checkpoints.zip, checkpoints.zip.zip or checkpoints.zip.ZIP.
/content/photo_restoration
/content/photo_restoration/Global
--2022-11-09 21:01:49--  https://facevc.blob.core.windows.net/zhanbo/old_photo/pretrain/Global/checkpoints.zip
Resolving facevc.blob.core.windows.net (facevc.blob.core.windows.net)... 20.60.228.1
Connecting to facevc.blob.core.windows.net (facevc.blob.core.windows.net)|20.60.228.1|:443... connected.
HTTP request sent, awaiting response... 404 The specified resource does not exist.
2022-11-09 21:01:50 ERROR 404: The specified resource does not exist..

unzip:  cannot find or open checkpoints.zip, checkpoints.zip.zip or checkpoints.zip.ZIP.
/content/photo_restoration

• added new GUI with options, settings, multilanguage and futuristic design.
• run.py now can be imported as library and accept parámeters, Is usted by the GUI.
• fixed error in run.py: error unknown argument > caused by argparse when the projets was executed, In a path that contain spaces or if input, output contain spaces. I added quotes '\"' to the input and output paths in all stages code.