main_pixelnerv2.py

import os
import warnings
warnings.filterwarnings("ignore")
import resource
rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
resource.setrlimit(resource.RLIMIT_NOFILE, (8192, rlimit[1]))

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
torch.set_float32_matmul_precision('medium')
torch.cuda.empty_cache()
torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
torch.multiprocessing.set_sharing_strategy('file_system')

from pytorch3d.renderer.cameras import (
    FoVPerspectiveCameras, 
    look_at_view_transform
)

from diffusers import UNet2DModel
# from pytorch_lightning.utilities.seed import seed_everything
from lightning_fabric.utilities.seed import seed_everything
from pytorch_lightning.callbacks import ModelCheckpoint
from pytorch_lightning.callbacks import LearningRateMonitor
from pytorch_lightning.callbacks import StochasticWeightAveraging
from pytorch_lightning.loggers import TensorBoardLogger
from pytorch_lightning import Trainer, LightningModule
from argparse import ArgumentParser
from typing import Optional
from monai.networks.nets import Unet, EfficientNetBN, DenseNet121
from monai.networks.layers.factories import Norm, Act
from monai.networks.layers import Reshape

from positional_encodings.torch_encodings import PositionalEncodingPermute3D
from datamodule import UnpairedDataModule
from dvr.renderer import (
    DirectVolumeFrontToBackRenderer, 
    DirectVolumeStratifiedFrontToBackRenderer,
    minimized, normalized, standardized
)

backbones = {
    "efficientnet-b0": (16, 24, 40, 112, 320),
    "efficientnet-b1": (16, 24, 40, 112, 320),
    "efficientnet-b2": (16, 24, 48, 120, 352),
    "efficientnet-b3": (24, 32, 48, 136, 384),
    "efficientnet-b4": (24, 32, 56, 160, 448),
    "efficientnet-b5": (24, 40, 64, 176, 512),
    "efficientnet-b6": (32, 40, 72, 200, 576),
    "efficientnet-b7": (32, 48, 80, 224, 640),
    "efficientnet-b8": (32, 56, 88, 248, 704),
    "efficientnet-l2": (72, 104, 176, 480, 1376),
}

class PixelNeRVFrontToBackFrustumFeaturer(nn.Module):
    def __init__(self, in_channels=1, out_channels=1, backbone="efficientnet-b7"):
        super().__init__()
        assert backbone in backbones.keys()
        self.model = EfficientNetBN(
            model_name=backbone, #(24, 32, 56, 160, 448)
            pretrained=True, 
            spatial_dims=2,
            in_channels=in_channels,
            num_classes=out_channels,
            adv_prop=True,
        )

    def forward(self, figures):
        camfeat = self.model.forward(figures)
        return camfeat

class PixelNeRVFrontToBackInverseRenderer(nn.Module):
    def __init__(self, in_channels=3, out_channels=1, shape=256, sh=0, pe=8, backbone="efficientnet-b7"):
        super().__init__()
        self.sh = sh
        self.pe = pe
        self.shape = shape
        assert backbone in backbones.keys()
        if self.pe>0:
            encoder_net = PositionalEncodingPermute3D(self.pe) # 8
            pe_channels = self.pe
            pos_enc = torch.ones([1, self.pe, self.shape, self.shape, self.shape])
            encoded = encoder_net(pos_enc)
            self.register_buffer('encoded', encoded)
        else:
            pe_channels = 0

        if self.sh > 0:
            from rsh import rsh_cart_2, rsh_cart_3
            # Generate grid
            zs = torch.linspace(-1, 1, steps=self.shape)
            ys = torch.linspace(-1, 1, steps=self.shape)
            xs = torch.linspace(-1, 1, steps=self.shape)
            z, y, x = torch.meshgrid(zs, ys, xs)
            zyx = torch.stack([z, y, x], dim=-1) # torch.Size([100, 100, 100, 3])
            if self.sh==2: 
                shw = rsh_cart_2(zyx.view(-1, 3)) 
                assert out_channels == 9
            elif self.sh==3: 
                shw = rsh_cart_3(zyx.view(-1, 3))
                assert out_channels == 16
            else:
                ValueError("Spherical Harmonics only support 2 and 3 degree")
            # self.register_buffer('shbasis', shw.unsqueeze(0).permute(0, 4, 1, 2, 3))
            self.register_buffer('shbasis', shw.view(out_channels, self.shape, self.shape, self.shape))
     
        self.clarity_net = UNet2DModel(
            sample_size=shape,  
            in_channels=1,  
            out_channels=shape,
            layers_per_block=2,  # how many ResNet layers to use per UNet block
            block_out_channels=(32, 48, 80, 224, 640), #(32, 48, 80, 224, 640),  # More channels -> more parameters
            norm_num_groups=8,
            down_block_types=(
                "DownBlock2D",  
                "DownBlock2D",  
                "DownBlock2D",
                "AttnDownBlock2D",  
                "AttnDownBlock2D",
            ),
            up_block_types=(
                "AttnUpBlock2D",
                "AttnUpBlock2D",    
                "UpBlock2D",
                "UpBlock2D",
                "UpBlock2D",    
            ),
            class_embed_type="timestep",
        )

        self.density_net = nn.Sequential(
            Unet(
                spatial_dims=3,
                in_channels=1+pe_channels,
                out_channels=1,
                channels=(32, 48, 80, 224, 640), #(32, 48, 80, 224, 640),
                strides=(2, 2, 2, 2, 2),
                num_res_units=2,
                kernel_size=3,
                up_kernel_size=3,
                act=("LeakyReLU", {"inplace": True}),
                norm=Norm.BATCH,
                # dropout=0.2,
            ),
        )

        self.mixture_net = nn.Sequential(
            Unet(
                spatial_dims=3,
                in_channels=2+pe_channels,
                out_channels=1,
                channels=(32, 48, 80, 224, 640), #(32, 48, 80, 224, 640),
                strides=(2, 2, 2, 2, 2),
                num_res_units=2,
                kernel_size=3,
                up_kernel_size=3,
                act=("LeakyReLU", {"inplace": True}),
                norm=Norm.BATCH,
                # dropout=0.2,
            ),
        )

        self.refiner_net = nn.Sequential(
            Unet(
                spatial_dims=3,
                in_channels=3+pe_channels,
                out_channels=out_channels,
                channels=(32, 48, 80, 224, 640), #(32, 48, 80, 224, 640),
                strides=(2, 2, 2, 2, 2),
                num_res_units=2,
                kernel_size=3,
                up_kernel_size=3,
                act=("LeakyReLU", {"inplace": True}),
                norm=Norm.BATCH,
                # dropout=0.2,
            ), 
        )
             
    def forward(self, figures, azim, elev, n_views=2):
        samples = figures.shape[0] # 3
        clarity = self.clarity_net(figures, azim*1000, elev*2000)[0].view(-1, 1, self.shape, self.shape, self.shape)
        
        # Multiview can stack along batch dimension, last dimension is for X-ray
        clarity_ct, clarity_xr = torch.split(clarity, n_views)
        clarity_ct = clarity_ct.mean(dim=0, keepdim=True)
        clarity = torch.cat([clarity_ct, clarity_xr])

        if self.pe > 0:
            density = self.density_net(torch.cat([self.encoded.repeat(clarity.shape[0], 1, 1, 1, 1), clarity], dim=1))
            mixture = self.mixture_net(torch.cat([self.encoded.repeat(clarity.shape[0], 1, 1, 1, 1), clarity, density], dim=1))
            shcoeff = self.refiner_net(torch.cat([self.encoded.repeat(clarity.shape[0], 1, 1, 1, 1), clarity, density, mixture], dim=1))
        else:
            density = self.density_net(torch.cat([clarity], dim=1))
            mixture = self.mixture_net(torch.cat([clarity, density], dim=1))
            shcoeff = self.refiner_net(torch.cat([clarity, density, mixture], dim=1))

        if self.sh > 0:
            # shcomps = shcoeff*self.shbasis.repeat(clarity.shape[0], 1, 1, 1, 1) 
            sh_comps_raw = torch.einsum('abcde,bcde->abcde', shcoeff, self.shbasis)
            # Take the absolute value of the spherical harmonic components
            sh_comps_abs = torch.abs(sh_comps_raw)
            sh_comps_max = sh_comps_abs.max()
            sh_comps_min = sh_comps_abs.min()
            # Normalize the spherical harmonic components
            shcomps = (sh_comps_abs - sh_comps_min) / (sh_comps_max - sh_comps_min + 1e-8)
        else:
            shcomps = shcoeff 

        volumes = torch.cat([clarity, shcomps], dim=1)
        volumes_ct, volumes_xr = torch.split(volumes, 1)
        volumes_ct = volumes_ct.repeat(n_views, 1, 1, 1, 1)
        volumes = torch.cat([volumes_ct, volumes_xr])
        return volumes 

def make_cameras(dist, elev, azim):
    assert dist.device == elev.device == azim.device
    _device = dist.device
    R, T = look_at_view_transform(
        dist=dist.float(), 
        elev=elev.float() * 90, 
        azim=azim.float() * 180
    )
    return FoVPerspectiveCameras(R=R, T=T, fov=45, aspect_ratio=1).to(_device)

class PixelNeRVLightningModule(LightningModule):
    def __init__(self, hparams, **kwargs):
        super().__init__()
        self.lr = hparams.lr
        self.gan = hparams.gan
        self.cam = hparams.cam
        self.shape = hparams.shape
        self.alpha = hparams.alpha
        self.gamma = hparams.gamma
        self.theta = hparams.theta
        self.omega = hparams.omega
        self.lambda_gp = hparams.lambda_gp
        self.clamp_val = hparams.clamp_val
       
        self.logsdir = hparams.logsdir
       
        self.st = hparams.st
        self.sh = hparams.sh
        self.pe = hparams.pe
        
        self.n_pts_per_ray = hparams.n_pts_per_ray
        self.weight_decay = hparams.weight_decay
        self.batch_size = hparams.batch_size
        self.backbone = hparams.backbone
        self.devices = hparams.devices
        
        self.save_hyperparameters()

        self.fwd_renderer = DirectVolumeStratifiedFrontToBackRenderer(
            image_width=self.shape, 
            image_height=self.shape, 
            n_pts_per_ray=self.n_pts_per_ray, 
            min_depth=2.0, 
            max_depth=6.0, 
        )
        
        self.inv_renderer = PixelNeRVFrontToBackInverseRenderer(
            in_channels=2, 
            out_channels=9 if self.sh==2 else 16 if self.sh==3 else 1, 
            shape=self.shape, 
            sh=self.sh, 
            pe=self.pe,
            backbone=self.backbone,
        )

        self.cam_settings = PixelNeRVFrontToBackFrustumFeaturer(
            in_channels=1, 
            out_channels=2, # azim + elev + prob
            backbone=self.backbone,
        )
        
        self.cam_settings.model._fc.weight.data.zero_()
        self.cam_settings.model._fc.bias.data.zero_()

        if self.gan:
            self.critic_model = PixelNeRVFrontToBackFrustumFeaturer(
                in_channels=1, 
                out_channels=1, # Bx1x16x16
                backbone=self.backbone,
            )
            # self.critic_model.model._fc.weight.data.zero_()
            # self.critic_model.model._fc.bias.data.zero_()

        self.loss = nn.L1Loss(reduction="mean")

    def forward_screen(self, image3d, cameras, stratified_sampling=False):   
        return self.fwd_renderer(image3d, cameras, stratified_sampling=stratified_sampling) 

    def forward_volume(self, image2d, azim, elev, n_views=2):      
        return self.inv_renderer(image2d * 2.0 - 1.0, azim.squeeze(), elev.squeeze(), n_views) 

    def forward_camera(self, image2d):
        return self.cam_settings(image2d * 2.0 - 1.0)

    def forward_critic(self, image2d):
        return self.critic_model(image2d * 2.0 - 1.0)

    def _common_step(self, batch, batch_idx, optimizer_idx, stage: Optional[str] = 'evaluation'):
        _device = batch["image3d"].device
        image3d = batch["image3d"]
        image2d = batch["image2d"]
            
        # Construct the random cameras
        # src_azim_random = torch.randn(self.batch_size, device=_device).clamp_(-0.9, 0.9)
        # src_elev_random = torch.randn(self.batch_size, device=_device).clamp_(-0.9, 0.9)
        src_azim_random = torch.distributions.uniform.Uniform(-1, 1).sample([self.batch_size]).to(_device) 
        src_elev_random = torch.distributions.uniform.Uniform(-1, 1).sample([self.batch_size]).to(_device) 
        src_dist_random = 4.0 * torch.ones(self.batch_size, device=_device)
        camera_random = make_cameras(src_dist_random, src_elev_random, src_azim_random)

        with torch.no_grad():
            est_figure_ct_random = self.forward_screen(image3d=image3d, cameras=camera_random, stratified_sampling=False)
            src_figure_xr_hidden = image2d

        est_dist_random = 4.0 * torch.ones(self.batch_size, device=_device)
        est_dist_hidden = 4.0 * torch.ones(self.batch_size, device=_device)
        est_dist_hidden = 4.0 * torch.ones(self.batch_size, device=_device)
        
        # Reconstruct the cameras
        est_feat_random, \
        est_feat_hidden = torch.split(
            self.forward_camera(
                image2d=torch.cat([est_figure_ct_random, src_figure_xr_hidden])
            ), self.batch_size
        )

        est_azim_random, est_elev_random = torch.split(est_feat_random, 1, dim=1)
        est_azim_hidden, est_elev_hidden = torch.split(est_feat_hidden, 1, dim=1)

        camera_random = make_cameras(est_dist_random, est_elev_random, est_azim_random)
        camera_hidden = make_cameras(est_dist_hidden, est_elev_hidden, est_azim_hidden)

        with torch.no_grad():
            est_figure_ct_hidden = self.forward_screen(image3d=image3d, cameras=camera_hidden, stratified_sampling=False)

        cam_view = [self.batch_size, 1]       
        # Jointly estimate the volumes, single view, random view and multiple views
        rng_figure = torch.randint(low=0, high=3, size=(1, 1))
        if stage=='train' and rng_figure==1:
            est_volume_ct_random, \
            est_volume_xr_hidden = torch.split(
                self.forward_volume(
                    image2d=torch.cat([est_figure_ct_random, src_figure_xr_hidden]),
                    azim=torch.cat([est_azim_random.view(cam_view), est_azim_hidden.view(cam_view)]),
                    elev=torch.cat([est_elev_random.view(cam_view), est_elev_hidden.view(cam_view)]),
                    n_views=1
                ), self.batch_size
            )
            est_volume_ct_hidden = est_volume_ct_random
        elif stage=='train' and rng_figure==2:
            est_volume_ct_hidden, \
            est_volume_xr_hidden = torch.split(
                self.forward_volume(
                    image2d=torch.cat([est_figure_ct_hidden, src_figure_xr_hidden]),
                    azim=torch.cat([est_azim_hidden.view(cam_view), est_azim_hidden.view(cam_view)]),
                    elev=torch.cat([est_elev_hidden.view(cam_view), est_elev_hidden.view(cam_view)]),
                    n_views=1
                ), self.batch_size
            )
            est_volume_ct_random = est_volume_ct_hidden
        else:
            est_volume_ct_random, \
            est_volume_ct_hidden, \
            est_volume_xr_hidden = torch.split(
                self.forward_volume(
                    image2d=torch.cat([est_figure_ct_random, est_figure_ct_hidden, src_figure_xr_hidden]),
                    azim=torch.cat([est_azim_random.view(cam_view), est_azim_hidden.view(cam_view), est_azim_hidden.view(cam_view)]),
                    elev=torch.cat([est_elev_random.view(cam_view), est_elev_hidden.view(cam_view), est_elev_hidden.view(cam_view)]),
                    n_views=2,
                ), self.batch_size
            )    
           
        # Reconstruct the appropriate XR
        rec_figure_ct_random = self.forward_screen(image3d=est_volume_ct_random[:,1:], cameras=camera_random, stratified_sampling=(stage=='train'))
        rec_figure_ct_hidden = self.forward_screen(image3d=est_volume_ct_hidden[:,1:], cameras=camera_hidden, stratified_sampling=(stage=='train'))
        est_figure_xr_hidden = self.forward_screen(image3d=est_volume_xr_hidden[:,1:], cameras=camera_hidden, stratified_sampling=(stage=='train'))

        # Perform Post activation like DVGO      
        mid_volume_ct_random = est_volume_ct_random[:,:1]
        mid_volume_ct_hidden = est_volume_ct_hidden[:,:1]
        mid_volume_xr_hidden = est_volume_xr_hidden[:,:1]

        est_volume_ct_random = est_volume_ct_random[:,1:].sum(dim=1, keepdim=True)
        est_volume_ct_hidden = est_volume_ct_hidden[:,1:].sum(dim=1, keepdim=True)
        est_volume_xr_hidden = est_volume_xr_hidden[:,1:].sum(dim=1, keepdim=True)

        # Compute the loss
        # Per-pixel_loss
        im2d_loss_ct_random = self.loss(est_figure_ct_random, rec_figure_ct_random) 
        im2d_loss_ct_hidden = self.loss(est_figure_ct_hidden, rec_figure_ct_hidden) 
        im2d_loss_xr_hidden = self.loss(src_figure_xr_hidden, est_figure_xr_hidden) 

        im3d_loss_ct_random = self.loss(image3d, est_volume_ct_random) + self.loss(image3d, mid_volume_ct_random) 
        im3d_loss_ct_hidden = self.loss(image3d, est_volume_ct_hidden) + self.loss(image3d, mid_volume_ct_hidden) 
    
        view_loss_ct_random = self.loss(torch.cat([src_azim_random, src_elev_random]), 
                                        torch.cat([est_azim_random, est_elev_random]))

        im2d_loss_ct = im2d_loss_ct_random + im2d_loss_ct_hidden 
        im2d_loss_xr = im2d_loss_xr_hidden
        im3d_loss_ct = im3d_loss_ct_random + im3d_loss_ct_hidden
        view_loss_ct = view_loss_ct_random 

        view_cond_xr = self.loss(est_azim_hidden, torch.zeros_like(est_azim_hidden)) \
                     + self.loss(est_elev_hidden, torch.zeros_like(est_elev_hidden))

        im2d_loss = im2d_loss_ct + im2d_loss_xr
        im3d_loss = im3d_loss_ct
        
        view_loss = view_loss_ct 
        view_cond = view_cond_xr

        self.log(f'{stage}_im2d_loss', im2d_loss, on_step=(stage=='train'), prog_bar=True, logger=True, sync_dist=True, batch_size=self.batch_size)
        self.log(f'{stage}_im3d_loss', im3d_loss, on_step=(stage=='train'), prog_bar=True, logger=True, sync_dist=True, batch_size=self.batch_size)
        self.log(f'{stage}_view_loss', view_loss, on_step=(stage=='train'), prog_bar=True, logger=True, sync_dist=True, batch_size=self.batch_size)
        p_loss = self.alpha*im3d_loss + self.theta*view_loss + self.gamma*im2d_loss + self.omega*view_cond

        if self.gan:
            if optimizer_idx==0:
                # Compute generator loss
                fake_images = torch.cat([rec_figure_ct_random, rec_figure_ct_hidden, est_figure_xr_hidden])
                fake_scores = self.forward_critic(fake_images)
                # g_loss = -torch.mean(fake_scores)
                g_loss = F.softplus(-fake_scores).mean()
                loss = p_loss + g_loss
                self.log(f'{stage}_g_loss', g_loss, on_step=(stage=='train'), prog_bar=False, logger=True, sync_dist=True, batch_size=self.batch_size)
            
            elif optimizer_idx==1:
                # Clamp parameters to enforce Lipschitz constraint
                for p in self.critic_model.parameters():
                    p.data.clamp_(-self.clamp_val, self.clamp_val)
                
                # Compute discriminator loss
                real_images = torch.cat([est_figure_ct_random, est_figure_ct_hidden, src_figure_xr_hidden])
                real_scores = self.forward_critic(real_images)
                fake_images = torch.cat([rec_figure_ct_random, rec_figure_ct_hidden, est_figure_xr_hidden])
                fake_scores = self.forward_critic(fake_images.detach())

                d_loss = F.softplus(-real_scores).mean() + F.softplus(+fake_scores).mean()
                loss = d_loss 
                self.log(f'{stage}_d_loss', d_loss, on_step=(stage=='train'), prog_bar=False, logger=True, sync_dist=True, batch_size=self.batch_size)
                
            else:
                loss = p_loss
        else:
            loss = p_loss

        if batch_idx==0:
            viz2d = torch.cat([
                        torch.cat([image3d[..., self.shape//2, :], 
                                   est_figure_ct_random,
                                   est_figure_ct_hidden,
                                   ], dim=-2).transpose(2, 3),
                        torch.cat([est_volume_ct_hidden[..., self.shape//2, :],
                                   rec_figure_ct_random,
                                   rec_figure_ct_hidden,
                                   ], dim=-2).transpose(2, 3),
                        torch.cat([image2d, 
                                   est_volume_xr_hidden[..., self.shape//2, :],
                                   est_figure_xr_hidden,
                                   ], dim=-2).transpose(2, 3),
                    ], dim=-2)
            grid = torchvision.utils.make_grid(viz2d, normalize=False, scale_each=False, nrow=1, padding=0)
            tensorboard = self.logger.experiment
            tensorboard.add_image(f'{stage}_samples', grid.clamp(0., 1.), self.current_epoch*self.batch_size + batch_idx)

        info = {f'loss': loss}
        return info

    # def training_step(self, batch, batch_idx):
    #     return self._common_step(batch, batch_idx, optimizer_idx=0, stage='train')
    def training_step(self, batch, batch_idx, optimizer_idx):
        return self._common_step(batch, batch_idx, optimizer_idx, stage='train')

    def validation_step(self, batch, batch_idx):
        return self._common_step(batch, batch_idx, optimizer_idx=-1, stage='validation')

    def test_step(self, batch, batch_idx):
        return self._common_step(batch, batch_idx, optimizer_idx=-1, stage='test')

    def _common_epoch_end(self, outputs, stage: Optional[str] = 'common'):
        loss = torch.stack([x[f'loss'] for x in outputs]).mean()
        self.log(f'{stage}_loss_epoch', loss, on_step=False, prog_bar=True, logger=True, sync_dist=True)

    def train_epoch_end(self, outputs):
        return self._common_epoch_end(outputs, stage='train')

    def validation_epoch_end(self, outputs):
        return self._common_epoch_end(outputs, stage='validation')

    def test_epoch_end(self, outputs):
        return self._common_epoch_end(outputs, stage='test')

    def configure_optimizers(self):
        opt_gen = torch.optim.AdamW([
            {'params': self.inv_renderer.parameters()},
            {'params': self.cam_settings.parameters()}
        ], lr=self.lr, betas=(0.5, 0.999))
        opt_dis = torch.optim.AdamW([
            {'params': self.critic_model.parameters()},
        ], lr=self.lr, betas=(0.5, 0.999))
        sch_gen = torch.optim.lr_scheduler.MultiStepLR(opt_gen, milestones=[100, 200], gamma=0.1)
        sch_dis = torch.optim.lr_scheduler.MultiStepLR(opt_dis, milestones=[100, 200], gamma=0.1)
        return [opt_gen, opt_dis], [sch_gen, sch_dis]
        # optimizer = torch.optim.AdamW(self.parameters(), lr=self.lr, betas=(0.5, 0.999))
        # scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[100, 200], gamma=0.1)
        # return [optimizer], [scheduler]

if __name__ == "__main__":
    parser = ArgumentParser()
    parser.add_argument("--conda_env", type=str, default="Unet")
    parser.add_argument("--notification_email", type=str, default="quantm88@gmail.com")

    # Model arguments
    parser.add_argument("--n_pts_per_ray", type=int, default=512, help="Sampling points per ray")
    parser.add_argument("--batch_size", type=int, default=1, help="size of the batches")
    parser.add_argument("--shape", type=int, default=256, help="spatial size of the tensor")
    parser.add_argument("--epochs", type=int, default=301, help="number of epochs")
    parser.add_argument("--train_samples", type=int, default=1000, help="training samples")
    parser.add_argument("--val_samples", type=int, default=400, help="validation samples")
    parser.add_argument("--test_samples", type=int, default=400, help="test samples")
    parser.add_argument("--st", type=int, default=1, help="with spatial transformer network")
    parser.add_argument("--sh", type=int, default=0, help="degree of spherical harmonic (2, 3)")
    parser.add_argument("--pe", type=int, default=0, help="positional encoding (0 - 8)")
    
    parser.add_argument("--gan", action="store_true", help="whether to train with GAN")
    parser.add_argument("--cam", action="store_true", help="train cam locked or hidden")
    parser.add_argument("--amp", action="store_true", help="train with mixed precision or not")
    
    parser.add_argument("--alpha", type=float, default=1., help="vol loss")
    parser.add_argument("--gamma", type=float, default=1., help="img loss")
    parser.add_argument("--theta", type=float, default=1., help="cam loss")
    parser.add_argument("--omega", type=float, default=1., help="cam cond")
    parser.add_argument("--lambda_gp", type=float, default=10, help="gradient penalty")
    parser.add_argument("--clamp_val", type=float, default=.1, help="gradient discrim clamp value")
    
    parser.add_argument("--lr", type=float, default=2e-4, help="adam: learning rate")
    parser.add_argument("--ckpt", type=str, default=None, help="path to checkpoint")
    parser.add_argument("--logsdir", type=str, default='logsfrecaling', help="logging directory")
    parser.add_argument("--datadir", type=str, default='data', help="data directory")
    parser.add_argument("--backbone", type=str, default='efficientnet-b7', help="Backbone for network")
    parser.add_argument("--weight_decay", type=float, default=1e-4, help="Weight decay")
    
    parser = Trainer.add_argparse_args(parser)

    # Collect the hyper parameters
    hparams = parser.parse_args()

    # Seed the application
    seed_everything(42)

    # Callback
    checkpoint_callback = ModelCheckpoint(
        dirpath=hparams.logsdir,
        # filename='epoch={epoch}-validation_loss={validation_loss_epoch:.2f}',
        monitor="validation_loss_epoch",
        auto_insert_metric_name=True, 
        save_top_k=-1,
        save_last=True,
        every_n_epochs=10,
    )
    lr_callback = LearningRateMonitor(logging_interval='step')

    # Logger
    tensorboard_logger = TensorBoardLogger(save_dir=hparams.logsdir, log_graph=True)
    swa_callback = StochasticWeightAveraging(swa_lrs=1e-2)
    # Init model with callbacks
    trainer = Trainer.from_argparse_args(
        hparams,
        max_epochs=hparams.epochs,
        logger=[tensorboard_logger],
        callbacks=[
            lr_callback,
            checkpoint_callback,
            # swa_callback
        ],
        accumulate_grad_batches=4,
        # strategy="ddp_sharded", #"horovod", #"deepspeed", #"ddp_sharded",
        strategy="ddp_sharded",  # "colossalai", "fsdp", #"ddp_sharded", #"horovod", #"deepspeed", #"ddp_sharded",
        # plugins=DDPStrategy(find_unused_parameters=False),
        precision=16 if hparams.amp else 32,
        # gradient_clip_val=0.01, 
        # gradient_clip_algorithm="value"
        # stochastic_weight_avg=True,
        # deterministic=False,
        # profiler="simple",
    )

    # Create data module
    train_image3d_folders = [
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/NSCLC/processed/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-0'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-1'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-2'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-3'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-4'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/Imagenglab/processed/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MELA2022/raw/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MELA2022/raw/val/images'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/AMOS2022/raw/train/images'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/AMOS2022/raw/val/images'),

        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/train/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/train/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/val/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/val/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/test/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/test/rawdata/'),

        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/UWSpine/processed/train/images'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/UWSpine/processed/test/images/'),
    ]

    train_label3d_folders = [
    ]

    train_image2d_folders = [
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/JSRT/processed/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/ChinaSet/processed/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/Montgomery/processed/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/VinDr/v1/processed/train/images/'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/VinDr/v1/processed/test/images/'),

        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/T62020/20200501/raw/images'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/T62021/20211101/raw/images'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/VinDr/v1/processed/train/images/'),
        # # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/VinDr/v1/processed/test/images/'),
    ]

    train_label2d_folders = [
    ]

    val_image3d_folders = [
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/NSCLC/processed/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-0'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-1'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-2'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-3'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MOSMED/processed/train/images/CT-4'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/Imagenglab/processed/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MELA2022/raw/train/images'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/MELA2022/raw/val/images'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/AMOS2022/raw/train/images'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/AMOS2022/raw/val/images'),

        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/train/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/train/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/val/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/val/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2019/raw/test/rawdata/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/Verse2020/raw/test/rawdata/'),

        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/UWSpine/processed/train/images'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/UWSpine/processed/test/images/'),
    ]

    val_image2d_folders = [
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/JSRT/processed/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/ChinaSet/processed/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/Montgomery/processed/images/'),
        # os.path.join(hparams.datadir, 'ChestXRLungSegmentation/VinDr/v1/processed/train/images/'),
        os.path.join(hparams.datadir, 'ChestXRLungSegmentation/VinDr/v1/processed/test/images/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/T62020/20200501/raw/images'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/T62021/20211101/raw/images'),
        # # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/VinDr/v1/processed/train/images/'),
        # os.path.join(hparams.datadir, 'SpineXRVertSegmentation/VinDr/v1/processed/test/images/'),
    ]

    test_image3d_folders = val_image3d_folders
    test_image2d_folders = val_image2d_folders

    datamodule = UnpairedDataModule(
        train_image3d_folders=train_image3d_folders,
        train_image2d_folders=train_image2d_folders,
        val_image3d_folders=val_image3d_folders,
        val_image2d_folders=val_image2d_folders,
        test_image3d_folders=test_image3d_folders,
        test_image2d_folders=test_image2d_folders,
        train_samples=hparams.train_samples,
        val_samples=hparams.val_samples,
        test_samples=hparams.test_samples,
        batch_size=hparams.batch_size,
        img_shape=hparams.shape,
        vol_shape=hparams.shape
    )
    datamodule.setup()

    ####### Test camera mu and bandwidth ########
    # test_random_uniform_cameras(hparams, datamodule)
    #############################################

    model = PixelNeRVLightningModule(
        hparams=hparams
    )
    # model = model.load_from_checkpoint(hparams.ckpt, strict=False) if hparams.ckpt is not None else model

    trainer.fit(
        model,
        datamodule,
        ckpt_path=hparams.ckpt if hparams.ckpt is not None else None, # "some/path/to/my_checkpoint.ckpt"
    )

    # test

    # serve