test.py

#
# Copyright (C) 2023, Inria
# GRAPHDECO research group, https://team.inria.fr/graphdeco
# All rights reserved.
#
# This software is free for non-commercial, research and evaluation use 
# under the terms of the LICENSE.md file.
#
# For inquiries contact  george.drettakis@inria.fr
#

import os
import numpy as np

import subprocess
# cmd = 'nvidia-smi -q -d Memory |grep -A4 GPU|grep Used'
# result = subprocess.run(cmd, shell=True, stdout=subprocess.PIPE).stdout.decode().split('\n')
# os.environ['CUDA_VISIBLE_DEVICES']=str(np.argmin([int(x.split()[2]) for x in result[:-1]]))

# os.system('echo $CUDA_VISIBLE_DEVICES')


import torch
import torchvision
import json
import wandb
import time
from os import makedirs
import shutil, pathlib
from pathlib import Path
from PIL import Image
import torchvision.transforms.functional as tf
# from lpipsPyTorch import lpips
# import lpips
from random import randint
from utils.loss_utils import l1_loss, ssim
from gaussian_renderer import prefilter_voxel, render, network_gui
import sys
from scene import Scene, GaussianModel
from utils.general_utils import safe_state
import uuid
from tqdm import tqdm
from utils.image_utils import psnr
from argparse import ArgumentParser, Namespace
from arguments import ModelParams, PipelineParams, OptimizationParams
from utils.encodings import anchor_round_digits, Q_anchor, get_binary_vxl_size
# torch.set_num_threads(32)


import lpips
lpips_fn = lpips.LPIPS(net='vgg').to('cuda')
# import pyiqa
# lpips = pyiqa.create_metric('lpips', as_loss=False).to("cuda")

bit2MB_scale = 8 * 1024 * 1024
run_codec = True

try:
    from torch.utils.tensorboard import SummaryWriter
    TENSORBOARD_FOUND = True
    print("found tf board")
except ImportError:
    TENSORBOARD_FOUND = False
    print("not found tf board")

def saveRuntimeCode(dst: str) -> None:
    additionalIgnorePatterns = ['.git', '.gitignore']
    ignorePatterns = set()
    ROOT = '.'
    with open(os.path.join(ROOT, '.gitignore')) as gitIgnoreFile:
        for line in gitIgnoreFile:
            if not line.startswith('#'):
                if line.endswith('\n'):
                    line = line[:-1]
                if line.endswith('/'):
                    line = line[:-1]
                ignorePatterns.add(line)
    ignorePatterns = list(ignorePatterns)
    for additionalPattern in additionalIgnorePatterns:
        ignorePatterns.append(additionalPattern)

    log_dir = pathlib.Path(__file__).parent.resolve()


    shutil.copytree(log_dir, dst, ignore=shutil.ignore_patterns(*ignorePatterns))

    print('Backup Finished!')



def render_set(model_path, name, iteration, views, gaussians, pipeline, background):
    render_path = os.path.join(model_path, name, "ours_{}".format(iteration), "renders")
    error_path = os.path.join(model_path, name, "ours_{}".format(iteration), "errors")
    gts_path = os.path.join(model_path, name, "ours_{}".format(iteration), "gt")

    makedirs(render_path, exist_ok=True)
    makedirs(error_path, exist_ok=True)
    makedirs(gts_path, exist_ok=True)

    t_list = []
    visible_count_list = []
    name_list = []
    per_view_dict = {}
    psnr_list = []
    for idx, view in enumerate(tqdm(views, desc="Rendering progress")):

        torch.cuda.synchronize(); t_start = time.time()
        voxel_visible_mask = prefilter_voxel(view, gaussians, pipeline, background)
        render_pkg = render(view, gaussians, pipeline, background, visible_mask=voxel_visible_mask)
        torch.cuda.synchronize(); t_end = time.time()
        torch.cuda.empty_cache()

        t_list.append(t_end - t_start)

        # renders
        rendering = torch.clamp(render_pkg["render"], 0.0, 1.0)
        visible_count = (render_pkg["radii"] > 0).sum()
        visible_count_list.append(visible_count)

        # gts
        gt = view.original_image[0:3, :, :]

        #
        gt_image = torch.clamp(view.original_image.to("cuda"), 0.0, 1.0)
        render_image = torch.clamp(rendering.to("cuda"), 0.0, 1.0)
        psnr_view = psnr(render_image, gt_image).mean().double()
        psnr_list.append(psnr_view)

        # error maps
        errormap = (rendering - gt).abs()


        name_list.append('{0:05d}'.format(idx) + ".png")
        torchvision.utils.save_image(rendering, os.path.join(render_path, '{0:05d}'.format(idx) + ".png"))
        torchvision.utils.save_image(errormap, os.path.join(error_path, '{0:05d}'.format(idx) + ".png"))
        torchvision.utils.save_image(gt, os.path.join(gts_path, '{0:05d}'.format(idx) + ".png"))
        per_view_dict['{0:05d}'.format(idx) + ".png"] = visible_count.item()

    with open(os.path.join(model_path, name, "ours_{}".format(iteration), "per_view_count.json"), 'w') as fp:
            json.dump(per_view_dict, fp, indent=True)

    print('testing_float_psnr=:', sum(psnr_list) / len(psnr_list))

    return t_list, visible_count_list


def render_sets(args_param, dataset : ModelParams, iteration : int, pipeline : PipelineParams, skip_train=True, skip_test=False, wandb=None, tb_writer=None, dataset_name=None, logger=None):        
    with torch.no_grad():
        gaussians = GaussianModel(
            dataset.feat_dim,
            dataset.n_offsets,
            dataset.voxel_size,
            dataset.update_depth,
            dataset.update_init_factor,
            dataset.update_hierachy_factor,
            dataset.use_feat_bank,
            n_features_per_level=args_param.n_features,
            decoded_version=False,
            level_num=args_param.level_num,
            hyper_divisor=args_param.hyper_divisor,
            target_ratio=args_param.target_ratio,
        )
        
        scene = Scene(dataset, gaussians, load_iteration=iteration, shuffle=False)

        log_info = scene.gaussians.estimate_final_bits()
        logger.info(log_info)
        if run_codec:  # conduct encoding and decoding
            bit_stream_path = os.path.join(args_param.model_path, 'bitstreams')
            os.makedirs(bit_stream_path, exist_ok=True)
            # conduct encoding
            log_info = scene.gaussians.conduct_encoding(pre_path_name=bit_stream_path)
            logger.info(log_info)
            # conduct decoding
            log_info = scene.gaussians.conduct_decoding(pre_path_name=bit_stream_path)
            logger.info(log_info)
            torch.cuda.empty_cache()

        gaussians.eval()
        
        bg_color = [1,1,1] if dataset.white_background else [0, 0, 0]
        background = torch.tensor(bg_color, dtype=torch.float32, device="cuda")

        if not skip_train:
            t_train_list, _  = render_set(dataset.model_path, "train", scene.loaded_iter, scene.getTrainCameras(), gaussians, pipeline, background)
            train_fps = 1.0 / torch.tensor(t_train_list[5:]).mean()
            logger.info(f'Train FPS: \033[1;35m{train_fps.item():.5f}\033[0m')
            if wandb is not None:
                wandb.log({"train_fps":train_fps.item(), })

        if not skip_test:
            t_test_list, visible_count = render_set(dataset.model_path, "test", scene.loaded_iter, scene.getTestCameras(), gaussians, pipeline, background)
            test_fps = 1.0 / torch.tensor(t_test_list[5:]).mean()
            logger.info(f'Test FPS: \033[1;35m{test_fps.item():.5f}\033[0m')
            if tb_writer:
                tb_writer.add_scalar(f'{dataset_name}/test_FPS', test_fps.item(), 0)
            if wandb is not None:
                wandb.log({"test_fps":test_fps, })

    return visible_count


def readImages(renders_dir, gt_dir):
    renders = []
    gts = []
    image_names = []
    for fname in os.listdir(renders_dir):
        render = Image.open(renders_dir / fname)
        gt = Image.open(gt_dir / fname)
        renders.append(tf.to_tensor(render).unsqueeze(0)[:, :3, :, :].cuda())
        gts.append(tf.to_tensor(gt).unsqueeze(0)[:, :3, :, :].cuda())
        image_names.append(fname)
    return renders, gts, image_names


def evaluate(model_paths, visible_count=None, wandb=None, tb_writer=None, dataset_name=None, logger=None):

    full_dict = {}
    per_view_dict = {}
    full_dict_polytopeonly = {}
    per_view_dict_polytopeonly = {}
    print("")

    scene_dir = model_paths
    full_dict[scene_dir] = {}
    per_view_dict[scene_dir] = {}
    full_dict_polytopeonly[scene_dir] = {}
    per_view_dict_polytopeonly[scene_dir] = {}

    test_dir = Path(scene_dir) / "test"

    for method in os.listdir(test_dir):

        full_dict[scene_dir][method] = {}
        per_view_dict[scene_dir][method] = {}
        full_dict_polytopeonly[scene_dir][method] = {}
        per_view_dict_polytopeonly[scene_dir][method] = {}

        method_dir = test_dir / method
        gt_dir = method_dir/ "gt"
        renders_dir = method_dir / "renders"
        renders, gts, image_names = readImages(renders_dir, gt_dir)

        ssims = []
        psnrs = []
        lpipss = []

        for idx in tqdm(range(len(renders)), desc=f"Metric evaluation progress [{method}]"):
            ssims.append(ssim(renders[idx], gts[idx]))
            psnrs.append(psnr(renders[idx], gts[idx]))
            lpipss.append(lpips_fn(renders[idx], gts[idx]).detach())

        if wandb is not None:
            wandb.log({"test_SSIMS":torch.stack(ssims).mean().item(), })
            wandb.log({"test_PSNR_final":torch.stack(psnrs).mean().item(), })
            wandb.log({"test_LPIPS":torch.stack(lpipss).mean().item(), })

        logger.info(f"model_paths: \033[1;35m{model_paths}\033[0m")
        logger.info("  SSIM : \033[1;35m{:>12.7f}\033[0m".format(torch.tensor(ssims).mean(), ".5"))
        logger.info("  PSNR : \033[1;35m{:>12.7f}\033[0m".format(torch.tensor(psnrs).mean(), ".5"))
        logger.info("  LPIPS: \033[1;35m{:>12.7f}\033[0m".format(torch.tensor(lpipss).mean(), ".5"))
        print("")


        if tb_writer:
            tb_writer.add_scalar(f'{dataset_name}/SSIM', torch.tensor(ssims).mean().item(), 0)
            tb_writer.add_scalar(f'{dataset_name}/PSNR', torch.tensor(psnrs).mean().item(), 0)
            tb_writer.add_scalar(f'{dataset_name}/LPIPS', torch.tensor(lpipss).mean().item(), 0)

            tb_writer.add_scalar(f'{dataset_name}/VISIBLE_NUMS', torch.tensor(visible_count).mean().item(), 0)

        full_dict[scene_dir][method].update({"SSIM": torch.tensor(ssims).mean().item(),
                                                "PSNR": torch.tensor(psnrs).mean().item(),
                                                "LPIPS": torch.tensor(lpipss).mean().item()})
        per_view_dict[scene_dir][method].update({"SSIM": {name: ssim for ssim, name in zip(torch.tensor(ssims).tolist(), image_names)},
                                                    "PSNR": {name: psnr for psnr, name in zip(torch.tensor(psnrs).tolist(), image_names)},
                                                    "LPIPS": {name: lp for lp, name in zip(torch.tensor(lpipss).tolist(), image_names)},
                                                    "VISIBLE_COUNT": {name: vc for vc, name in zip(torch.tensor(visible_count).tolist(), image_names)}})

    with open(scene_dir + "/results.json", 'w') as fp:
        json.dump(full_dict[scene_dir], fp, indent=True)
    with open(scene_dir + "/per_view.json", 'w') as fp:
        json.dump(per_view_dict[scene_dir], fp, indent=True)

def get_logger(path):
    import logging

    logger = logging.getLogger()
    logger.setLevel(logging.INFO)
    fileinfo = logging.FileHandler(os.path.join(path, "outputs.log"))
    fileinfo.setLevel(logging.INFO)
    controlshow = logging.StreamHandler()
    controlshow.setLevel(logging.INFO)
    formatter = logging.Formatter("%(asctime)s - %(levelname)s: %(message)s")
    fileinfo.setFormatter(formatter)
    controlshow.setFormatter(formatter)

    logger.addHandler(fileinfo)
    logger.addHandler(controlshow)

    return logger

if __name__ == "__main__":
    # Set up command line argument parser
    parser = ArgumentParser(description="Training script parameters")
    lp = ModelParams(parser)
    op = OptimizationParams(parser)
    pp = PipelineParams(parser)
    parser.add_argument('--level_num', type=int, default=3)
    parser.add_argument('--level_scale', type=int, default=10)
    parser.add_argument('--ip', type=str, default="127.0.0.1")
    parser.add_argument('--port', type=int, default=6009)
    parser.add_argument('--debug_from', type=int, default=-1)
    parser.add_argument('--detect_anomaly', action='store_true', default=False)
    parser.add_argument('--warmup', action='store_true', default=False)
    parser.add_argument('--use_wandb', action='store_true', default=False)
    # parser.add_argument("--test_iterations", nargs="+", type=int, default=[11_000, 15_000, 20_000, 25_000, 29_000, 30_000])
    parser.add_argument("--test_iterations", nargs="+", type=int, default=[30_000])
    # parser.add_argument("--save_iterations", nargs="+", type=int, default=[11_000, 15_000, 20_000, 25_000, 29_000, 30_000])
    parser.add_argument("--save_iterations", nargs="+", type=int, default=[30_000])
    parser.add_argument("--quiet", action="store_true")
    parser.add_argument("--checkpoint_iterations", nargs="+", type=int, default=[])
    parser.add_argument("--start_checkpoint", type=str, default = None)
    parser.add_argument("--gpu", type=str, default = '-1')
    parser.add_argument("--log2", type=int, default = 13)
    parser.add_argument("--log2_2D", type=int, default = 15)
    parser.add_argument("--n_features", type=int, default = 4)
    parser.add_argument("--disable_hyper", default=False, action="store_true")
    args = parser.parse_args(sys.argv[1:])
    args.save_iterations.append(args.iterations)


    # enable logging

    model_path = args.model_path
    os.makedirs(model_path, exist_ok=True)

    logger = get_logger(model_path)


    logger.info(f'args: {args}')

    if args.gpu != '-1':
        os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu)
        os.system("echo $CUDA_VISIBLE_DEVICES")
        logger.info(f'using GPU {args.gpu}')

    '''try:
        saveRuntimeCode(os.path.join(args.model_path, 'backup'))
    except:
        logger.info(f'save code failed~')'''

    dataset = args.source_path.split('/')[-2]
    exp_name = args.model_path.split('/')[-1]

    if args.use_wandb:
        wandb.login()
        run = wandb.init(
            # Set the project where this run will be logged
            project=f"Ours-{dataset}",
            name=exp_name,
            # Track hyperparameters and run metadata
            settings=wandb.Settings(start_method="fork"),
            config=vars(args)
        )
    else:
        wandb = None

    logger.info("Optimizing " + args.model_path)

    # Initialize system state (RNG)
    safe_state(args.quiet)

    # Start GUI server, configure and run training
    args.port = np.random.randint(10000, 20000)
    # network_gui.init(args.ip, args.port)
    torch.autograd.set_detect_anomaly(args.detect_anomaly)

    # rendering
    logger.info(f'\nStarting Rendering~')
    visible_count = render_sets(args, lp.extract(args), -1, pp.extract(args), wandb=wandb, logger=logger)
    logger.info("\nRendering complete.")

    # calc metrics
    logger.info("\n Starting evaluation...")
    evaluate(args.model_path, visible_count=visible_count, wandb=wandb, logger=logger)
    logger.info("\nEvaluating complete.")