How to reproduce the results of the paper?

[mochan-shs]

Thanks for your released code.

When conducting experiments on the Horse scene, Francis scene and Church scene of the Tanks dataset using the default.yaml file, I found that the results were much worse than those in the paper. PSNR, SSIM were worse than those in the paper. How can I set up the yaml file to get better results? The default.yaml was configured as follow:

model:
  num_layers: 8
  freeze_network: False
  use_image_feature: False
  network_type: official
  occ_activation: softplus
  hidden_dim: 256
  pos_enc_levels: 10
  dir_enc_levels: 4
dataloading:
  dataset_name: any
  path: 
  scene: []
  batchsize: 1
  n_workers: 1
  img_size:
  path:
  with_depth: False
  with_mask: False
  spherify: True
  customized_poses: False #use poses other than colmap
  customized_focal: False #use focal other than colmap
  resize_factor: 
  depth_net: dpt
  crop_size: 0
  random_ref: 1
  norm_depth: False
  load_colmap_poses: True
  shuffle: True
  sample_rate: 8

rendering:
  type: nope_nerf # changed
  n_max_network_queries: 64000
  white_background: False
  radius: 4.0
  num_points: 128
  depth_range: [0.01, 10]
  dist_alpha: False
  use_ray_dir: True
  normalise_ray: True
  normal_loss: False
  sample_option: uniform
  outside_steps: 0
depth:
  type: None
  path: DPT/dpt_hybrid-midas-501f0c75.pt
  non_negative: True
  scale: 0.000305
  shift: 0.1378
  invert: True
  freeze: True
pose:
  learn_pose: True
  learn_R: True
  learn_t: True
  init_pose: False
  init_R_only: False
  learn_focal: False
  update_focal: True
  fx_only: False
  focal_order: 2
  init_pose_type: gt
  init_focal_type: gt
distortion:
  learn_distortion: True
  fix_scaleN: True
  learn_scale: True
  learn_shift: True
training:
  type: nope_nerf
  load_dir: model.pt
  load_pose_dir: model_pose.pt
  load_focal_dir: model_focal.pt
  load_distortion_dir: model_distortion.pt
  n_training_points: 1024
  scheduling_epoch: 10000
  batch_size: 1
  learning_rate: 0.001
  focal_lr: 0.001
  pose_lr: 0.0005
  distortion_lr: 0.0005
  weight_decay: 0.0
  scheduler_gamma_pose: 0.9
  scheduler_gamma: 0.9954
  scheduler_gamma_distortion: 0.9
  scheduler_gamma_focal: 0.9
  validate_every: -1
  visualize_every: 10000
  eval_pose_every: 1 # epoch
  eval_img_every: 1 # epoch
  print_every: 100
  backup_every: 10000
  checkpoint_every: 5000
  rgb_weight: [1.0, 1.0]
  depth_weight: [0.04, 0.0]
  weight_dist_2nd_loss: [0.0, 0.0]
  weight_dist_1st_loss: [0.0, 0.0]
  pc_weight: [1.0, 0.0]
  rgb_s_weight: [1.0, 0.0]
  depth_consistency_weight: [0.0, 0.0]
  rgb_loss_type: l1
  depth_loss_type: l1
  log_scale_shift_per_view: False
  with_auto_mask: False
  vis_geo: True
  vis_resolution: [54, 96]
  mode: train
  with_ssim: False
  use_gt_depth: False
  load_ckpt_model_only: False 
  optim: Adam
  detach_gt_depth: False
  match_method: dense
  pc_ratio: 4
  shift_first: False
  detach_ref_img: True
  scheduling_start: 10000
  auto_scheduler: True
  length_smooth: 1000
  patient: 30
  scale_pcs: True
  detach_rgbs_scale: False
  scheduling_mode: 
  vis_reprojection_every: 5000
  nearest_limit: 0.01
  annealing_epochs: 2000 #should be >=0
extract_images:
  extraction_dir: extraction
  N_novel_imgs: 120
  traj_option: bspline
  use_learnt_poses: True
  use_learnt_focal: True
  resolution:
  model_file: model.pt
  model_file_pose: model_pose.pt
  model_file_focal: model_focal.pt
  eval_depth: False
  bspline_degree: 100
eval_pose:
  n_points: 1024
  type: nope_nerf
  type_to_eval: eval
  opt_pose_epoch: 1000
  extraction_dir: extraction
  init_method: pre
  opt_eval_lr: 0.001

[Yuhuoo]

I have encountered the same problem. The results I have obtained are worse than those in the paper, but all results were achieved without modifying the code and config.

eval:

python evaluation/eval.py configs/Tanks/Ignatius.yaml
Mean MSE: 0.00, PSNR: 23.96, SSIM: 0.62, LPIPS 0.47
23.96 &0.62 & 0.47

python evaluation/eval.py configs/Tanks/Horse.yaml
Mean MSE: 0.00, PSNR: 24.37, SSIM: 0.74, LPIPS 0.38
24.37 &0.74 & 0.38

python evaluation/eval.py configs/Tanks/Family.yaml
Mean MSE: 0.00, PSNR: 23.23, SSIM: 0.66, LPIPS 0.50
23.23 &0.66 & 0.50

python evaluation/eval.py configs/Tanks/Museum.yaml
Mean MSE: 0.00, PSNR: 25.76, SSIM: 0.75, LPIPS 0.36
25.76 &0.75 & 0.36

python evaluation/eval.py configs/Tanks/Ballroom.yaml
Mean MSE: 0.00, PSNR: 24.14, SSIM: 0.68, LPIPS 0.42
24.14 &0.68 & 0.42

python evaluation/eval.py configs/Tanks/Barn.yaml
Mean MSE: 0.00, PSNR: 25.59, SSIM: 0.67, LPIPS 0.46
25.59 &0.67 & 0.46

python evaluation/eval.py configs/Tanks/Francis.yaml
Mean MSE: 0.00, PSNR: 26.38, SSIM: 0.73, LPIPS 0.47
26.38 &0.73 & 0.47

eval_pose:

python evaluation/eval_poses.py configs/Tanks/Ignatius.yaml
RPE_trans:0.037 & RPE_rot:0.008 & ATE:0.004

python evaluation/eval_poses.py configs/Tanks/Horse.yaml
RPE_trans:1.301 & RPE_rot:0.158 & ATE:0.026

python evaluation/eval_poses.py configs/Tanks/Family.yaml
RPE_trans:0.097 & RPE_rot:0.064 & ATE:0.014

python evaluation/eval_poses.py configs/Tanks/Museum.yaml
RPE_trans:0.510 & RPE_rot:0.316 & ATE:0.017

python evaluation/eval_poses.py configs/Tanks/Ballroom.yaml
RPE_trans:0.098 & RPE_rot:0.042 & ATE:0.006

python evaluation/eval_poses.py configs/Tanks/Barn.yaml
RPE_trans:0.182 & RPE_rot:0.040 & ATE:0.022

python evaluation/eval.py configs/Tanks/Francis.yaml
RPE_trans:0.482 & RPE_rot:0.208 & ATE:0.042