run_nerf_many_generalize_sample.py

import os, sys
import numpy as np
import imageio
import json
import random
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from tqdm import tqdm, trange
from os.path import join
import matplotlib.pyplot as plt

from run_nerf_helpers import *

from load_generalized import load_many_data, get_train_ids, get_test_ids

from multiplane_helpers_generalized import MultiImageNeRF, ImagePlane
from dataset import NeRFShapeNetDataset
from torch.utils.data import DataLoader

import pandas as pd


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
np.random.seed(0)
DEBUG = False

def calculate_image_metrics(entry, render_kwargs, metric_fn, count, args):
    x = []
    y = []
    with torch.no_grad():
        for c in range(count):
            img_i = np.random.choice(len(entry['images']), 1)
            target = entry['images'][img_i][0].to(device)
            target = torch.Tensor(target.float())
            pose = entry['cam_poses'][img_i, :3,:4][0].to(device)
            
            H = entry["images"].shape[1]
            W = entry["images"].shape[2]
            focal = .5 * W / np.tan(.5 * 0.6911112070083618) 

            K = np.array([
            [focal, 0, 0.5*W],
            [0, focal, 0.5*H],
            [0, 0, 1]
            ])

            img, _, _, _ = render(H, W, K, chunk=args.netchunk, c2w=pose,
                                                        verbose=True, retraw=True,
                                                        **render_kwargs)
            
            x.append(img)
            y.append(target)
        
        x = torch.stack(x)
        y = torch.stack(y)

        metric_val = metric_fn(y, x)

        return metric_val

def batchify(fn, chunk):
    """Constructs a version of 'fn' that applies to smaller batches.
    """
    if chunk is None:
        return fn
    def ret(inputs):
        return torch.cat([fn(inputs[i:i+chunk]) for i in range(0, inputs.shape[0], chunk)], 0)
    return ret


def run_network(inputs, viewdirs, fn, embed_fn, embeddirs_fn, netchunk=1024*64):
    """Prepares inputs and applies network 'fn'.
    """
    inputs_flat = torch.reshape(inputs, [-1, inputs.shape[-1]])
    embedded = embed_fn(inputs_flat)

    if viewdirs is not None:
        input_dirs = viewdirs[:,None].expand(inputs.shape)
        input_dirs_flat = torch.reshape(input_dirs, [-1, input_dirs.shape[-1]])
        embedded_dirs = embeddirs_fn(input_dirs_flat)
        embedded = torch.cat([embedded, embedded_dirs], -1)

    outputs_flat = batchify(fn, netchunk)(embedded)
    outputs = torch.reshape(outputs_flat, list(inputs.shape[:-1]) + [outputs_flat.shape[-1]])
    return outputs


def batchify_rays(rays_flat, chunk=1024*32, **kwargs):
    """Render rays in smaller minibatches to avoid OOM.
    """
    all_ret = {}
    for i in range(0, rays_flat.shape[0], chunk):
        ret = render_rays(rays_flat[i:i+chunk], **kwargs)
        for k in ret:
            if k not in all_ret:
                all_ret[k] = []
            all_ret[k].append(ret[k])

    all_ret = {k : torch.cat(all_ret[k], 0) for k in all_ret}
    return all_ret


def render(H, W, K, chunk=1024*32, rays=None, c2w=None, ndc=True,
                  near=0., far=1.,
                  use_viewdirs=False, c2w_staticcam=None,
                  **kwargs):
    """Render rays
    Args:
      H: int. Height of image in pixels.
      W: int. Width of image in pixels.
      focal: float. Focal length of pinhole camera.
      chunk: int. Maximum number of rays to process simultaneously. Used to
        control maximum memory usage. Does not affect final results.
      rays: array of shape [2, batch_size, 3]. Ray origin and direction for
        each example in batch.
      c2w: array of shape [3, 4]. Camera-to-world transformation matrix.
      ndc: bool. If True, represent ray origin, direction in NDC coordinates.
      near: float or array of shape [batch_size]. Nearest distance for a ray.
      far: float or array of shape [batch_size]. Farthest distance for a ray.
      use_viewdirs: bool. If True, use viewing direction of a point in space in model.
      c2w_staticcam: array of shape [3, 4]. If not None, use this transformation matrix for 
       camera while using other c2w argument for viewing directions.
    Returns:
      rgb_map: [batch_size, 3]. Predicted RGB values for rays.
      disp_map: [batch_size]. Disparity map. Inverse of depth.
      acc_map: [batch_size]. Accumulated opacity (alpha) along a ray.
      extras: dict with everything returned by render_rays().
    """
    if c2w is not None:
        # special case to render full image
        rays_o, rays_d = get_rays(H, W, K, c2w)
    else:
        # use provided ray batch
        rays_o, rays_d = rays

    if use_viewdirs:
        # provide ray directions as input
        viewdirs = rays_d
        if c2w_staticcam is not None:
            # special case to visualize effect of viewdirs
            rays_o, rays_d = get_rays(H, W, K, c2w_staticcam)
        viewdirs = viewdirs / torch.norm(viewdirs, dim=-1, keepdim=True)
        viewdirs = torch.reshape(viewdirs, [-1,3]).float()

    sh = rays_d.shape # [..., 3]
    if ndc:
        # for forward facing scenes
        rays_o, rays_d = ndc_rays(H, W, K[0][0], 1., rays_o, rays_d)

    # Create ray batch
    rays_o = torch.reshape(rays_o, [-1,3]).float()
    rays_d = torch.reshape(rays_d, [-1,3]).float()

    near, far = near * torch.ones_like(rays_d[...,:1]), far * torch.ones_like(rays_d[...,:1])
    rays = torch.cat([rays_o, rays_d, near, far], -1)
    if use_viewdirs:
        rays = torch.cat([rays, viewdirs], -1)

    # Render and reshape
    all_ret = batchify_rays(rays, chunk, **kwargs)
    for k in all_ret:
        k_sh = list(sh[:-1]) + list(all_ret[k].shape[1:])
        all_ret[k] = torch.reshape(all_ret[k], k_sh)

    k_extract = ['rgb_map', 'disp_map', 'acc_map']
    ret_list = [all_ret[k] for k in k_extract]
    ret_dict = {k : all_ret[k] for k in all_ret if k not in k_extract}
    return ret_list + [ret_dict]


def render_path(render_poses, hwf, K, chunk, render_kwargs, gt_imgs=None, savedir=None, render_factor=0):

    H, W, focal = hwf

    if render_factor!=0:
        # Render downsampled for speed
        H = H//render_factor
        W = W//render_factor
        focal = focal/render_factor

    rgbs = []
    disps = []

    t = time.time()
    for i, c2w in enumerate(tqdm(render_poses)):
        print(i, time.time() - t)
        t = time.time()
        rgb, disp, acc, _ = render(H, W, K, chunk=chunk, c2w=c2w[:3,:4], **render_kwargs)
        rgbs.append(rgb.cpu().numpy())
        disps.append(disp.cpu().numpy())

        """
        if gt_imgs is not None and render_factor==0:
            p = -10. * np.log10(np.mean(np.square(rgb.cpu().numpy() - gt_imgs[i])))
            print(p)
        """

        if savedir is not None:
            rgb8 = rgb.cpu().numpy() #to8b(rgbs[-1])
            filename = os.path.join(savedir, '{:03d}.png'.format(i))
            imageio.imwrite(filename, rgb8)

    
    rgbs = np.stack(rgbs, 0)
    
    if gt_imgs is not None:
        p = -10. * np.log10(np.mean(np.square(rgbs - gt_imgs.cpu().numpy())))
        print(" CALCULATED PSNR FOR TESTSET")
        print(p)
        
    disps = np.stack(disps, 0)

    return rgbs, disps, p


def create_mi_nerf(count, args):
    """Instantiate NeRF's MLP model.
    """
    embed_fn, input_ch = get_embedder(args.multires, args.i_embed)

    input_ch_views = 0
    embeddirs_fn = None
    if args.use_viewdirs:
        embeddirs_fn, input_ch_views = get_embedder(args.multires_views, args.i_embed)
    output_ch = 5 if args.N_importance > 0 else 4
    
    # Use MultiImage
    model = MultiImageNeRF(None, count, input_ch_views).to(device)
    grad_vars = list(model.parameters())

    network_query_fn = lambda inputs, viewdirs, network_fn : run_network(inputs, viewdirs, network_fn,
                                                                embed_fn=embed_fn,
                                                                embeddirs_fn=embeddirs_fn,
                                                                netchunk=args.netchunk)
    
    model_fine = None
    if args.N_importance > 0:
        model_fine = MultiImageNeRF(None, count, input_ch_views).to(device)
        grad_vars += list(model_fine.parameters())

    # Create optimizer
    #optimizer = torch.optim.Adam(params=grad_vars, lr=args.lrate, betas=(0.9, 0.999))

    start = 0
    basedir = args.basedir
    expname = args.expname

    ##########################

    # Load checkpoints
    if args.ft_path is not None and args.ft_path!='None':
        ckpts = [args.ft_path]
    else:
        ckpts = [os.path.join(basedir, expname, f) for f in sorted(os.listdir(os.path.join(basedir, expname))) if 'tar' in f]

    print('Found ckpts', ckpts)
    if len(ckpts) > 0 and not args.no_reload:
        ckpt_path = ckpts[-1]
        print('Reloading from', ckpt_path)
        ckpt = torch.load(ckpt_path)

        start = ckpt['global_step']
        #optimizer.load_state_dict(ckpt['optimizer_state_dict'])

        # Load model
        model.load_state_dict(ckpt['network_fn_state_dict'])
        model_fine.load_state_dict(ckpt['network_fine_state_dict'])


    ##########################

    render_kwargs_train = {
        'network_query_fn' : network_query_fn,
        'perturb' : args.perturb,
        'N_importance' : args.N_importance,
        'network_fine' : model_fine,
        'N_samples' : args.N_samples,
        'network_fn' : model,
        'use_viewdirs' : args.use_viewdirs,
        'white_bkgd' : args.white_bkgd,
        'raw_noise_std' : args.raw_noise_std,
    }

    # NDC only good for LLFF-style forward facing data
    if args.dataset_type != 'llff' or args.no_ndc:
        print('Not ndc!')
        render_kwargs_train['ndc'] = False
        render_kwargs_train['lindisp'] = args.lindisp

    render_kwargs_test = {k : render_kwargs_train[k] for k in render_kwargs_train}
    render_kwargs_test['perturb'] = False
    render_kwargs_test['raw_noise_std'] = 0.

    return render_kwargs_train, render_kwargs_test, start, grad_vars#, optimizer

def raw2outputs(raw, z_vals, rays_d, raw_noise_std=0, white_bkgd=False, pytest=False):
    """Transforms model's predictions to semantically meaningful values.
    Args:
        raw: [num_rays, num_samples along ray, 4]. Prediction from model.
        z_vals: [num_rays, num_samples along ray]. Integration time.
        rays_d: [num_rays, 3]. Direction of each ray.
    Returns:
        rgb_map: [num_rays, 3]. Estimated RGB color of a ray.
        disp_map: [num_rays]. Disparity map. Inverse of depth map.
        acc_map: [num_rays]. Sum of weights along each ray.
        weights: [num_rays, num_samples]. Weights assigned to each sampled color.
        depth_map: [num_rays]. Estimated distance to object.
    """
    raw2alpha = lambda raw, dists, act_fn=F.relu: 1.-torch.exp(-act_fn(raw)*dists)

    dists = z_vals[...,1:] - z_vals[...,:-1]
    dists = torch.cat([dists, torch.Tensor([1e10]).expand(dists[...,:1].shape)], -1)  # [N_rays, N_samples]

    dists = dists * torch.norm(rays_d[...,None,:], dim=-1)

    rgb = torch.sigmoid(raw[...,:3])  # [N_rays, N_samples, 3]
    noise = 0.
    if raw_noise_std > 0.:
        noise = torch.randn(raw[...,3].shape) * raw_noise_std

        # Overwrite randomly sampled data if pytest
        if pytest:
            np.random.seed(0)
            noise = np.random.rand(*list(raw[...,3].shape)) * raw_noise_std
            noise = torch.Tensor(noise)

    alpha = raw2alpha(raw[...,3] + noise, dists)  # [N_rays, N_samples]
    # weights = alpha * tf.math.cumprod(1.-alpha + 1e-10, -1, exclusive=True)
    weights = alpha * torch.cumprod(torch.cat([torch.ones((alpha.shape[0], 1)), 1.-alpha + 1e-10], -1), -1)[:, :-1]
    rgb_map = torch.sum(weights[...,None] * rgb, -2)  # [N_rays, 3]

    depth_map = torch.sum(weights * z_vals, -1)
    disp_map = 1./torch.max(1e-10 * torch.ones_like(depth_map), depth_map / torch.sum(weights, -1))
    acc_map = torch.sum(weights, -1)

    if white_bkgd:
        rgb_map = rgb_map + (1.-acc_map[...,None])

    return rgb_map, disp_map, acc_map, weights, depth_map


def render_rays(ray_batch,
                network_fn,
                network_query_fn,
                N_samples,
                retraw=False,
                lindisp=False,
                perturb=0.,
                N_importance=0,
                network_fine=None,
                white_bkgd=False,
                raw_noise_std=0.,
                verbose=False,
                pytest=False):
    """Volumetric rendering.
    Args:
      ray_batch: array of shape [batch_size, ...]. All information necessary
        for sampling along a ray, including: ray origin, ray direction, min
        dist, max dist, and unit-magnitude viewing direction.
      network_fn: function. Model for predicting RGB and density at each point
        in space.
      network_query_fn: function used for passing queries to network_fn.
      N_samples: int. Number of different times to sample along each ray.
      retraw: bool. If True, include model's raw, unprocessed predictions.
      lindisp: bool. If True, sample linearly in inverse depth rather than in depth.
      perturb: float, 0 or 1. If non-zero, each ray is sampled at stratified
        random points in time.
      N_importance: int. Number of additional times to sample along each ray.
        These samples are only passed to network_fine.
      network_fine: "fine" network with same spec as network_fn.
      white_bkgd: bool. If True, assume a white background.
      raw_noise_std: ...
      verbose: bool. If True, print more debugging info.
    Returns:
      rgb_map: [num_rays, 3]. Estimated RGB color of a ray. Comes from fine model.
      disp_map: [num_rays]. Disparity map. 1 / depth.
      acc_map: [num_rays]. Accumulated opacity along each ray. Comes from fine model.
      raw: [num_rays, num_samples, 4]. Raw predictions from model.
      rgb0: See rgb_map. Output for coarse model.
      disp0: See disp_map. Output for coarse model.
      acc0: See acc_map. Output for coarse model.
      z_std: [num_rays]. Standard deviation of distances along ray for each
        sample.
    """
    N_rays = ray_batch.shape[0]
    rays_o, rays_d = ray_batch[:,0:3], ray_batch[:,3:6] # [N_rays, 3] each
    viewdirs = ray_batch[:,-3:] if ray_batch.shape[-1] > 8 else None
    bounds = torch.reshape(ray_batch[...,6:8], [-1,1,2])
    near, far = bounds[...,0], bounds[...,1] # [-1,1]

    t_vals = torch.linspace(0., 1., steps=N_samples)
    if not lindisp:
        z_vals = near * (1.-t_vals) + far * (t_vals)
    else:
        z_vals = 1./(1./near * (1.-t_vals) + 1./far * (t_vals))

    z_vals = z_vals.expand([N_rays, N_samples])

    if perturb > 0.:
        # get intervals between samples
        mids = .5 * (z_vals[...,1:] + z_vals[...,:-1])
        upper = torch.cat([mids, z_vals[...,-1:]], -1)
        lower = torch.cat([z_vals[...,:1], mids], -1)
        # stratified samples in those intervals
        t_rand = torch.rand(z_vals.shape)

        # Pytest, overwrite u with numpy's fixed random numbers
        if pytest:
            np.random.seed(0)
            t_rand = np.random.rand(*list(z_vals.shape))
            t_rand = torch.Tensor(t_rand)

        z_vals = lower + (upper - lower) * t_rand

    pts = rays_o[...,None,:] + rays_d[...,None,:] * z_vals[...,:,None] # [N_rays, N_samples, 3]


#     raw = run_network(pts)
    raw = network_query_fn(pts, viewdirs, network_fn)
    rgb_map, disp_map, acc_map, weights, depth_map = raw2outputs(raw, z_vals, rays_d, raw_noise_std, white_bkgd, pytest=pytest)

    if N_importance > 0:

        rgb_map_0, disp_map_0, acc_map_0 = rgb_map, disp_map, acc_map

        z_vals_mid = .5 * (z_vals[...,1:] + z_vals[...,:-1])
        z_samples = sample_pdf(z_vals_mid, weights[...,1:-1], N_importance, det=(perturb==0.), pytest=pytest)
        z_samples = z_samples.detach()

        z_vals, _ = torch.sort(torch.cat([z_vals, z_samples], -1), -1)
        pts = rays_o[...,None,:] + rays_d[...,None,:] * z_vals[...,:,None] # [N_rays, N_samples + N_importance, 3]

        run_fn = network_fn if network_fine is None else network_fine
#         raw = run_network(pts, fn=run_fn)
        raw = network_query_fn(pts, viewdirs, run_fn)

        rgb_map, disp_map, acc_map, weights, depth_map = raw2outputs(raw, z_vals, rays_d, raw_noise_std, white_bkgd, pytest=pytest)

    ret = {'rgb_map' : rgb_map, 'disp_map' : disp_map, 'acc_map' : acc_map}
    if retraw:
        ret['raw'] = raw
    if N_importance > 0:
        ret['rgb0'] = rgb_map_0
        ret['disp0'] = disp_map_0
        ret['acc0'] = acc_map_0
        ret['z_std'] = torch.std(z_samples, dim=-1, unbiased=False)  # [N_rays]

    for k in ret:
        if (torch.isnan(ret[k]).any() or torch.isinf(ret[k]).any()) and DEBUG:
            print(f"! [Numerical Error] {k} contains nan or inf.")

    return ret


def config_parser():

    import configargparse
    parser = configargparse.ArgumentParser()
    parser.add_argument('--config', is_config_file=True, 
                        help='config file path')
    parser.add_argument("--expname", type=str, 
                        help='experiment name')
    parser.add_argument("--basedir", type=str, default='./logs/', 
                        help='where to store ckpts and logs')
    parser.add_argument("--datadir", type=str, default='./data/llff/fern', 
                        help='input data directory')

    # training options
    parser.add_argument("--netdepth", type=int, default=8, 
                        help='layers in network')
    parser.add_argument("--netwidth", type=int, default=256, 
                        help='channels per layer')
    parser.add_argument("--netdepth_fine", type=int, default=8, 
                        help='layers in fine network')
    parser.add_argument("--netwidth_fine", type=int, default=256, 
                        help='channels per layer in fine network')
    parser.add_argument("--N_rand", type=int, default=32*32*4, 
                        help='batch size (number of random rays per gradient step)')
    parser.add_argument("--lrate", type=float, default=1e-4, 
                        help='learning rate')
    parser.add_argument("--lrate_decay", type=int, default=250, 
                        help='exponential learning rate decay (in 1000 steps)')
    parser.add_argument("--chunk", type=int, default=1024*32, 
                        help='number of rays processed in parallel, decrease if running out of memory')
    parser.add_argument("--netchunk", type=int, default=1024*64, 
                        help='number of pts sent through network in parallel, decrease if running out of memory')
    parser.add_argument("--no_batching", action='store_true', 
                        help='only take random rays from 1 image at a time')
    parser.add_argument("--no_reload", action='store_true', 
                        help='do not reload weights from saved ckpt')
    parser.add_argument("--ft_path", type=str, default=None, 
                        help='specific weights npy file to reload for coarse network')

    # rendering options
    parser.add_argument("--N_samples", type=int, default=64, 
                        help='number of coarse samples per ray')
    parser.add_argument("--N_importance", type=int, default=0,
                        help='number of additional fine samples per ray')
    parser.add_argument("--perturb", type=float, default=1.,
                        help='set to 0. for no jitter, 1. for jitter')
    parser.add_argument("--use_viewdirs", action='store_true', 
                        help='use full 5D input instead of 3D')
    parser.add_argument("--i_embed", type=int, default=-1, 
                        help='set 0 for default positional encoding, -1 for none')
    parser.add_argument("--multires", type=int, default=10, 
                        help='log2 of max freq for positional encoding (3D location)')
    parser.add_argument("--multires_views", type=int, default=4, 
                        help='log2 of max freq for positional encoding (2D direction)')
    parser.add_argument("--raw_noise_std", type=float, default=0., 
                        help='std dev of noise added to regularize sigma_a output, 1e0 recommended')

    parser.add_argument("--render_only", action='store_true', 
                        help='do not optimize, reload weights and render out render_poses path')
    parser.add_argument("--render_test", action='store_true', 
                        help='render the test set instead of render_poses path')
    parser.add_argument("--render_factor", type=int, default=0, 
                        help='downsampling factor to speed up rendering, set 4 or 8 for fast preview')

    # training options
    parser.add_argument("--precrop_iters", type=int, default=0,
                        help='number of steps to train on central crops')
    parser.add_argument("--precrop_frac", type=float,
                        default=.5, help='fraction of img taken for central crops') 

    # dataset options
    parser.add_argument("--dataset_type", type=str, default='llff', 
                        help='options: llff / blender / deepvoxels')
    parser.add_argument("--testskip", type=int, default=8, 
                        help='will load 1/N images from test/val sets, useful for large datasets like deepvoxels')
    
    parser.add_argument("--dataset_sample", type=str, default='cars', 
                        help='')

    ## deepvoxels flags
    parser.add_argument("--shape", type=str, default='greek', 
                        help='options : armchair / cube / greek / vase')

    ## blender flags
    parser.add_argument("--white_bkgd", action='store_true', 
                        help='set to render synthetic data on a white bkgd (always use for dvoxels)')
    parser.add_argument("--half_res", action='store_true', 
                        help='load blender synthetic data at 400x400 instead of 800x800')

    ## llff flags
    parser.add_argument("--factor", type=int, default=8, 
                        help='downsample factor for LLFF images')
    parser.add_argument("--no_ndc", action='store_true', 
                        help='do not use normalized device coordinates (set for non-forward facing scenes)')
    parser.add_argument("--lindisp", action='store_true', 
                        help='sampling linearly in disparity rather than depth')
    parser.add_argument("--spherify", action='store_true', 
                        help='set for spherical 360 scenes')
    parser.add_argument("--llffhold", type=int, default=8, 
                        help='will take every 1/N images as LLFF test set, paper uses 8')

    # logging/saving options
    parser.add_argument("--i_print",   type=int, default=100, 
                        help='frequency of console printout and metric loggin')
    parser.add_argument("--i_img",     type=int, default=500, 
                        help='frequency of tensorboard image logging')
    parser.add_argument("--i_weights", type=int, default=10000, 
                        help='frequency of weight ckpt saving')
    parser.add_argument("--i_testset", type=int, default=10000, 
                        help='frequency of testset saving')
    parser.add_argument("--i_video",   type=int, default=50000, 
                        help='frequency of render_poses video saving')

    return parser


def train():

    parser = config_parser()
    args = parser.parse_args()

    # Load data
    K = None

    if args.dataset_type == 'many':
        
        dataset = NeRFShapeNetDataset(root_dir='/home/spurek/hyperrf_data/ds', classes=[args.dataset_sample])
        dataloader = DataLoader(dataset, batch_size=4, shuffle=True, drop_last=True, pin_memory=False, generator=torch.Generator(device='cuda'))
        
        dataset_test = NeRFShapeNetDataset(root_dir='/home/spurek/hyperrf_data/ds', classes=[args.dataset_sample], train=False)
        dataloader_test = DataLoader(dataset_test, batch_size=1, shuffle=False, drop_last=True, pin_memory=False, generator=torch.Generator(device='cuda'))
        
        objects, test_objects, render_poses, hwf = load_many_data(f'/home/spurek/hyperrf_data/ds/{args.dataset_sample}')
        print('Loaded many')
        near = 2.
        far = 6.
        
    else:
        print('Unknown dataset type', args.dataset_type, 'exiting')
        return

    # Cast intrinsics to right types
    H, W, focal = hwf
    H, W = int(H), int(W)
    hwf = [H, W, focal]

    if K is None:
        K = np.array([
            [focal.item(), 0, 0.5*W],
            [0, focal.item(), 0.5*H],
            [0, 0, 1]
        ])

    # Create log dir and copy the config file
    basedir = args.basedir
    expname = args.expname
    os.makedirs(os.path.join(basedir, expname), exist_ok=True)
    f = os.path.join(basedir, expname, 'args.txt')
    with open(f, 'w') as file:
        for arg in sorted(vars(args)):
            attr = getattr(args, arg)
            file.write('{} = {}\n'.format(arg, attr))
    if args.config is not None:
        f = os.path.join(basedir, expname, 'config.txt')
        with open(f, 'w') as file:
            file.write(open(args.config, 'r').read())

    # Create nerf model
    render_kwargs_train, render_kwargs_test, start, grad_vars = create_mi_nerf(50, args)
    global_step = start
 
    bds_dict = {
        'near' : near,
        'far' : far,
    }
    render_kwargs_train.update(bds_dict)
    render_kwargs_test.update(bds_dict)

    # Move testing data to GPU
    render_poses = torch.Tensor(render_poses).to(device)

    objcs_for_training = list(objects.values())
    random.shuffle(objcs_for_training)
    for bi, batch in enumerate(dataloader_test):
        
        # Limit number of rendered objects
        #if bi >= 300:
        #    break
        
        results_dir = join(basedir, expname, 'examples_200', str(bi))
        os.makedirs(results_dir, exist_ok=True)
        for oi in range(batch['images'].size(0)):
            object = {'images': batch['images'][oi].float(), 'cam_poses': batch['cam_poses'][oi].float()}

            image_plane = ImagePlane(focal, object['cam_poses'].cpu().numpy(), object['images'].cpu().numpy(), 50)

            render_kwargs_test['network_fn'].image_plane = image_plane
            render_kwargs_test['network_fine'].image_plane = image_plane
            
            for gt in range(10):
                imageio.imsave(join(results_dir, f'ground_t_{gt}.png'), to8b(object['images'][gt].detach().cpu().numpy()))

                with torch.no_grad():
                    img_i = [gt]
                    target = object['images'][img_i][0].to(device)
                    target = torch.Tensor(target.float())
                    pose = object['cam_poses'][img_i, :3,:4][0].to(device)

                    img_r, _, _, _ = render(H, W, K, chunk=args.chunk, c2w = pose,
                                                                verbose=True, retraw=True,
                                                                **render_kwargs_test)

                    frame = torch.cat([img_r], dim=1)

                    imageio.imsave(join(results_dir, f'rec_{gt}.png'), to8b(frame.detach().cpu().numpy()))


if __name__=='__main__':
    torch.set_default_tensor_type('torch.cuda.FloatTensor')

    train()