| loss: |
| mse: |
| weight: 1.0 |
| lpips: |
| weight: 0.5 |
| apply_after_step: 0 |
| perceptual_loss: true |
| deltas: |
| weight: 1 |
| exclude_by_norm_grad: true |
| exclude_by_norm_grad_opposite: true |
| eps: 1.0e-08 |
| apply_after_step: 100 |
| dataset: |
| view_sampler: |
| name: boundedv2 |
| num_target_views: 6 |
| num_context_views: 8 |
| min_distance_between_context_views: 24 |
| max_distance_between_context_views: 45 |
| max_distance_to_context_views: 0 |
| context_gap_warm_up_steps: 10000 |
| target_gap_warm_up_steps: 0 |
| initial_min_distance_between_context_views: 20 |
| initial_max_distance_between_context_views: 30 |
| initial_max_distance_to_context_views: 0 |
| extra_views_sampling_strategy: farthest_point |
| target_views_replace_sample: false |
| name: dl3dv |
| roots: |
| - datasets/dl3dv-480p-chunks |
| make_baseline_1: false |
| augment: true |
| image_shape: |
| - 256 |
| - 448 |
| background_color: |
| - 0.0 |
| - 0.0 |
| - 0.0 |
| cameras_are_circular: false |
| baseline_epsilon: 0.001 |
| max_fov: 100.0 |
| skip_bad_shape: true |
| near: 0.01 |
| far: 200.0 |
| baseline_scale_bounds: false |
| shuffle_val: true |
| test_len: -1 |
| test_chunk_interval: 1 |
| sort_target_index: true |
| sort_context_index: true |
| train_times_per_scene: 1 |
| test_times_per_scene: 1 |
| ori_image_shape: |
| - 270 |
| - 480 |
| overfit_to_scene: null |
| overfit_max_views: 148 |
| use_index_to_load_chunk: false |
| mix_tartanair: false |
| no_mix_test_set: true |
| load_depth: false |
| center_pose: false |
| opencv_pose_format: true |
| min_views: 0 |
| max_views: 0 |
| highres: false |
| mix_re10k: false |
| re10k_min_view_dist: 40 |
| re10k_max_view_dist: 300 |
| pose_align_first_view: false |
| pose_align_middle_view: false |
| scale_extrinsics: 1.0 |
| metric_scale_align_dl3dv: false |
| load_remain_context: false |
| num_remain_context: 8 |
| random_crop: false |
| min_size: |
| - 384 |
| - 512 |
| max_size: |
| - 512 |
| - 960 |
| scene_trainer: |
| scene_initializer: |
| per_pixel: true |
| per_view: true |
| name: resplat |
| num_depth_candidates: 128 |
| num_surfaces: 1 |
| gaussians_per_pixel: 1 |
| gaussian_adapter: |
| gaussian_scale_min: 0.5 |
| gaussian_scale_max: 0.3 |
| sh_degree: 3 |
| exp_scale: false |
| softplus_scale: true |
| clamp_min_scale: 1.0e-06 |
| scale_detach_depth: false |
| exp_scale_bias: 4.0 |
| no_rotate_sh: true |
| no_sh_mask: true |
| init_rotation_identity: false |
| d_feature: 128 |
| visualizer: |
| num_samples: 8 |
| min_resolution: 256 |
| export_ply: false |
| unimatch_weights_path: pretrained/gmdepth-scale1-resumeflowthings-scannet-5d9d7964.pth |
| multiview_trans_attn_split: 2 |
| costvolume_unet_feat_dim: 128 |
| costvolume_unet_channel_mult: |
| - 1 |
| - 1 |
| - 1 |
| costvolume_unet_attn_res: [] |
| depth_unet_feat_dim: 64 |
| depth_unet_attn_res: [] |
| depth_unet_channel_mult: |
| - 1 |
| - 1 |
| - 1 |
| downscale_factor: 4 |
| shim_patch_size: 4 |
| local_mv_match: 2 |
| monodepth_vit_type: vits |
| supervise_intermediate_depth: true |
| return_depth: true |
| num_scales: 1 |
| upsample_factor: 8 |
| lowest_feature_resolution: 8 |
| depth_unet_channels: 128 |
| grid_sample_disable_cudnn: false |
| large_gaussian_head: false |
| color_large_unet: false |
| init_sh_input_img: true |
| feature_upsampler_channels: 64 |
| gaussian_regressor_channels: 256 |
| unet_gaussian_regressor: false |
| resnet_gaussian_regressor: false |
| train_depth_only: false |
| pt_head: true |
| pt_heads: 1 |
| init_pt_with_mv_attn: false |
| init_pt_with_mv_attn_lowres: false |
| pt_head_channels: null |
| pt_head_concat_img: false |
| pt_head_conv: false |
| multi_scale_pt: false |
| attn_proj_channels: 64 |
| fps_num_samples: null |
| knn_samples: 16 |
| post_norm: false |
| no_rpe: true |
| no_knn_attn: false |
| num_blocks: 4 |
| pt_downsample: 0 |
| fps_agg_func: attn |
| subsample_method: fps |
| add_pt_residual: true |
| pt_pred_residual_position: false |
| freeze_depth: false |
| use_gt_depth: false |
| separate_depth_color: false |
| separate_depth_type: small |
| separate_depth_gaussian_scale: false |
| sample_log_depth: true |
| bilinear_upsample_depth: false |
| no_upsample_depth: false |
| return_lowres_depth: false |
| foundationstereo: false |
| fstereo_num_refine: 1 |
| lvsm_gaussian_regressor: false |
| lvsm_layers: 6 |
| latent_gs: true |
| latent_downsample: 4 |
| fixed_latent_size: true |
| latent_gs_img_interp: area |
| dpt_head_depth: false |
| latent_dpt_upsampler: false |
| latent_dpt_upsampler_no_concat: false |
| light_dpt_feature: false |
| avgpool_depth: false |
| nearest_down_depth: false |
| predict_scale: false |
| norm_by_points: false |
| no_pred_depth_range: false |
| point_dist_init_gaussian_scale: false |
| resizeconv_upsampler: false |
| depth_pred_half_res: false |
| use_amp: true |
| pt_head_amp: true |
| use_fsdp: false |
| use_checkpointing: false |
| init_use_checkpointing: false |
| new_gaussian_order: true |
| rotate_quat_to_world: false |
| refine_rotate_quat_to_world: false |
| refine_no_use_covariance: false |
| latent_new_reshape: false |
| ptv3: false |
| ptv3_grid_size: 0.1 |
| no_pixel_offset: false |
| init_gaussian_multiple: 1 |
| deform_sample_depth: false |
| deform_sample_depth_debug: false |
| scene_optimizer: |
| refiner: |
| name: none |
| do_densify: false |
| do_prune: false |
| do_opacity_reset: false |
| cap_max: -1 |
| noise_lr: 0.0 |
| pause_refine_after_reset: 0 |
| refine_every: 999999999 |
| reset_every: 999999999 |
| refine_start_iter: 999999999 |
| refine_stop_iter: 999999999 |
| refine_scale2d_stop_iter: 0 |
| grow_grad2d: 0.0 |
| grow_scale3d: 0.0 |
| prune_scale3d: 0.0 |
| prune_scale2d: 0.0 |
| grow_scale2d: 0.0 |
| min_opacity: 0.0 |
| prune_zero_radii: false |
| reduce_opacity: false |
| reduce_factor: 0.0 |
| reduce_every: 0 |
| lr_scheduler: |
| lr_data: |
| _base: 1 |
| _means: 1 |
| _scales: 1 |
| _quats: 1 |
| _opacities: 1 |
| _sh0: 1 |
| _shN: 1 |
| apply_scheduler: |
| _base: false |
| _means: false |
| _scales: false |
| _quats: false |
| _opacities: false |
| _sh0: false |
| _shN: false |
| name: none |
| no_refine_mean: false |
| no_refine_scale: false |
| no_refine_rotation: false |
| no_refine_opacity: false |
| no_refine_sh0: false |
| no_refine_shN: false |
| name: clogs |
| no_render_error: false |
| refine_sh_only: false |
| num_basic_refine_blocks: 4 |
| num_refine_blocks: 1 |
| input_error_lpips_features: false |
| input_error_pool_vgg_features: false |
| input_error_use_all_vgg_features: false |
| input_error_vit_feature: false |
| input_error_resnet_feature: true |
| inpu_error_no_freeze_resnet_feature: false |
| input_error_shallow_resnet_feature: false |
| input_error_resnet_feature_layers: 18 |
| input_error_convnext_feature: false |
| input_error_convnext_feature_size: small |
| input_error_concat_feature: false |
| input_error_concat_feature_cosine: false |
| input_error_cosine_feature: false |
| input_error_add_feature: false |
| input_error_concat_rgb_feature_error: false |
| concat_init_state: false |
| replace_init_state: false |
| state_channels: 256 |
| refine_block_rmsnorm: false |
| refine_block_layernorm: false |
| pt_qk_norm: false |
| norm_pt_block: false |
| refine_gaussian_multiple: 1 |
| refine_residual_init_state: false |
| clamp_refine_max_scale: 3.0 |
| clamp_min_scale: 1.0e-06 |
| clamp_min_raw_opacities: -7 |
| clamp_max_raw_opacities: 7 |
| gaussian_head_multiple: 1 |
| update_head_concat_img: false |
| update_head_layer_num: 2 |
| update_head_act: gelu |
| update_head_final_act: identity |
| update_head_scale_mag: false |
| update_head_scalar_scale: true |
| update_head_scalar_scale_act: relu |
| update_head_hidden_dim_matches: "output" |
| update_attn_proj_channels: 64 |
| update_no_knn_attn: false |
| update_no_tran_block_norm: false |
| update_tran_block_act: gelu |
| multi_gaussian_scale_smaller: false |
| init_gaussian_multiple: 1 |
| refine_condition_pt_feature: true |
| input_error_no_abs: false |
| reinit_gaussian_when_refine_multiple: false |
| refine_same_num_points: false |
| input_error_no_shuffle: false |
| input_error_cache_resnet_feature: false |
| input_error_view_pool_resnet_feature: false |
| input_error_global_pool_resnet_feature: false |
| init_state_wo_features: false |
| init_state_type: constant |
| init_state_scale: 0 |
| pt_heads: 1 |
| refine_with_mv_attn: false |
| refine_with_mv_attn_lowres: false |
| refine_no_mv_attn: false |
| mv_attn_conv_with_norm: false |
| refine_mv_shuffle_attn: false |
| refine_mv_attn_with_pos_enc: false |
| refine_shuffle_attn_no_norm: false |
| refine_mv_unimatch_attn: false |
| refine_knn_samples: 16 |
| refine_multi_scale_pt: false |
| input_alpha: false |
| input_depth: false |
| input_depth_smooth_error: false |
| input_error: false |
| input_zero_gaussian: false |
| input_zero_state: false |
| input_zero_init_state: false |
| input_error_cross_attn: false |
| input_error_cross_attn_blocks: 1 |
| input_error_cross_attn_with_mlp: false |
| input_error_radii_averaged: false |
| input_error_additional_cross_attn: false |
| input_error_num_intermediate_views: 8 |
| input_error_mv_attn: false |
| input_error_mv_attn_blocks: 2 |
| input_error_mv_attn_lowres: false |
| input_error_mv_attn_proj_channels: 0 |
| input_error_lowres_attn_down_factor: 4 |
| input_error_lowres_attn_pos_enc: false |
| input_error_mv_attn_swin: false |
| input_error_num_views: 0 |
| input_error_remain_context: false |
| input_error_merge_remain_context: false |
| input_error_warp_remain_context: false |
| input_error_random_num_remain_context: false |
| input_error_num_remain_context_test: 0 |
| input_error_warp_input_view: false |
| window_local_refine: false |
| window_global_refine: false |
| window_local_global_refine: false |
| update_window_size: 0 |
| local_gaussian_render: false |
| input_error_half_res: false |
| input_gradient: true |
| input_gradient_log: false |
| input_gradient_log_clip_deltas: 0.001 |
| input_gradient_scale: 1.0 |
| gradient_update_scale: 1.0 |
| input_gradient_with_ssim_loss: true |
| input_gradient_same_loss: false |
| scale_residual_grads: false |
| train_min_refine: 1 |
| train_max_refine: 6 |
| local_global_update: false |
| num_global_update: 0 |
| train_global_update_only: false |
| random_update_with_size: false |
| ptv3: false |
| ptv3_grid_size: 0.1 |
| use_amp: true |
| pt_head_amp: true |
| pt_update_amp: true |
| use_checkpointing: false |
| recurrent_use_checkpointing: false |
| debug_refine_update_module: true |
| debug_refine_mlp: false |
| debug_refine_mlp_layers: 0 |
| debug_refine_mlp_fixed_init: false |
| input_gradient_normalize: true |
| input_gradient_normalize_type: adam |
| input_normalize_state: false |
| input_normalize_gaussians: false |
| predict_state_scale: true |
| opt_scales_before_act: false |
| scale_initial_opacities: 1.0 |
| sh_d: 16 |
| local_prune_zero_radii: false |
| local_prune_low_weights: false |
| local_prune_low_weights_thresh: -1 |
| update_only_nonzero_grad: true |
| experimental_run: false |
| experimental_update: |
| _base: true |
| _means: true |
| _scales: true |
| _quats: true |
| _opacities: true |
| _sh0: true |
| _shN: true |
| experimental_use_grads: false |
| experimental_use_norm_grads: |
| _base: false |
| _means: false |
| _scales: false |
| _quats: false |
| _opacities: false |
| _sh0: false |
| _shN: false |
| experimental_lr: |
| _base: 1 |
| _means: 0.00016 |
| _scales: 0.005 |
| _opacities: 0.05 |
| _quats: 0.001 |
| _sh0: 0.0025 |
| _shN: 0.000125 |
| sample_init_gaussians: 0 |
| use_time_encoding: false |
| time_encoding_max_steps: 2000 |
| decoder: |
| name: gsplat |
| use_covariances: false |
| use_fsdp: false |
| train_scene_init: false |
| train_scene_opt: true |
| num_update_steps: 4 |
| iter_batch_size: -1 |
| iter_batch_size_min: -1 |
| iter_batch_size_max: -1 |
| initializer: |
| num_depth_candidates: 128 |
| costvolume_unet_feat_dim: 128 |
| costvolume_unet_channel_mult: |
| - 1 |
| - 1 |
| - 1 |
| costvolume_unet_attn_res: |
| - 4 |
| gaussians_per_pixel: 1 |
| depth_unet_feat_dim: 32 |
| depth_unet_attn_res: |
| - 16 |
| depth_unet_channel_mult: |
| - 1 |
| - 1 |
| - 1 |
| - 1 |
| - 1 |
| shim_patch_size: 16 |
| test: |
| postprocessing: |
| steps: 2000 |
| compute_metrics_every: 100 |
| lr_data: |
| _base: 1 |
| _means: 1 |
| _scales: 1 |
| _opacities: 1 |
| _quats: 1 |
| _sh0: 1 |
| _shN: 1 |
| scheduler: null |
| scheduler_warm_up_ratio: 0.01 |
| __target__: src.scene_trainer.postprocessing.NoPostProcessCfg |
| name: none |
| output_path: null |
| compute_scores: true |
| compute_scores_metrics: [psnr,ssim,lpips] |
| eval_time_skip_steps: 0 |
| eval_initialization: true |
| save_render_image: false |
| save_gt_image: false |
| save_render_depth: false |
| save_gt_depth: false |
| save_error_image: false |
| save_error_depth: false |
| save_video: false |
| save_depth: false |
| save_depth_npy: false |
| save_depth_concat_img: false |
| save_gaussian: false |
| save_poses: false |
| no_align_to_view: false |
| save_point_cloud: false |
| render_chunk_size: null |
| stablize_camera: false |
| stab_camera_kernel: 50 |
| eval_context_views: false |
| inference_window_size: null |
| profile_model: false |
| save_colmap_train_test_views: false |
| ori_colmap_data_path: null |
| adam_optimizer_step: 0 |
| save_at_iters: null |
| save_every_freq: |
| - 1 |
| - 10 |
| - 100 |
| - 500 |
| save_every_steps: |
| - 0 |
| - 10 |
| - 100 |
| - 1000 |
| skip_if_outputs_exist: false |
| dec_chunk_size: 30 |
| wandb: |
| project: ablation-dl3dv-8views |
| entity: placeholder |
| name: dl3dv |
| mode: online |
| id: null |
| notes: '' |
| tags: |
| - dl3dv |
| - 270x480 |
| mode: train |
| data_loader: |
| train: |
| num_workers: 10 |
| persistent_workers: true |
| batch_size: 1 |
| seed: 1234 |
| test: |
| num_workers: 4 |
| persistent_workers: false |
| batch_size: 1 |
| seed: 2345 |
| val: |
| num_workers: 1 |
| persistent_workers: true |
| batch_size: 1 |
| seed: 3456 |
| meta_optimizer: |
| lr: 0.0001 |
| lr_monodepth: 0.0 |
| lr_depth: 0.0 |
| warm_up_steps: 2000 |
| weight_decay: 0.01 |
| warm_up_ratio: 0.01 |
| adamw_8bit: false |
| checkpointing: |
| load: null |
| every_n_train_steps: 1000 |
| save_top_k: 5 |
| pretrained_model: checkpoints/optgs/unified-dl3dv-8views/init/checkpoints/epoch_20-step_100000.ckpt |
| pretrained_model_rel_dir: ${checkpoint_rel_dir:${checkpointing.pretrained_model}} |
| pretrained_monodepth: null |
| pretrained_mvdepth: null |
| pretrained_depth: null |
| pretrained_scale_predictor: null |
| pretrained_depth_teacher: null |
| no_strict_load: true |
| resume: false |
| no_resume_upsampler: false |
| partial_load: false |
| freeze_mono_vit: false |
| freeze_mono_vit: false |
| resume_update_module: null |
| train: |
| depth_mode: null |
| extended_visualization: false |
| print_log_every_n_steps: 100 |
| eval_model_every_n_val: 2 |
| eval_data_length: 999999 |
| eval_deterministic: false |
| eval_time_skip_steps: 3 |
| eval_save_model: true |
| l1_loss: true |
| intermediate_loss_weight: 0.9 |
| no_viz_video: false |
| eval_depth: false |
| forward_depth_only: false |
| train_ignore_large_loss: 0.0 |
| no_log_projections: true |
| no_log_video: true |
| depth_loss_weight: 0.0 |
| log_depth_loss: true |
| depth_smooth_loss_weight: 0.0 |
| depth_smooth_loss_nonorm: false |
| depth_smooth_loss_weight_nvs: 0.0 |
| monodepth_loss_weight: 0.0 |
| depth_teacher_loss_weight: 0.0 |
| viz_depth_teacher: false |
| eval_render_depth: false |
| render_depth_loss_weight: 0.0 |
| viz_render_depth: false |
| use_gt_depth_range: false |
| depth_range_from_disparity: false |
| max_disparity: 128.0 |
| min_disparity: 4.0 |
| img_warp_loss_weight: 0.0 |
| warp_loss_start_step: 5000 |
| loss_on_input_views: true |
| loss_on_target_views: true |
| loss_on_input_views_num: 4 |
| train_window_size: null |
| half_res_lpips_loss: false |
| viz_depth_separate: false |
| use_replay_buffer: true |
| replay_buffer_cfg: |
| capacity: 20 |
| sample_batch_size: 1 |
| sample_prob: 0.7 |
| insert_prob: 0.99 |
| return_prob: 0.99 |
| simulate_ahead: true |
| simulate_ahead_min_steps: 1 |
| simulate_ahead_max_steps: 50 |
| simulate_ahead_grow: 10000 |
| max_t: null |
| push_only_if_not_full: false |
| seed: 111123 |
| meta_trainer: |
| max_steps: 100000 |
| val_check_interval: 0.25 |
| gradient_clip_val: 0.5 |
| num_sanity_val_steps: 2 |
| eval_index: null |
| limit_test_batches: 1.0 |
| limit_train_batches: 1.0 |
| num_nodes: 1 |
| output_dir: !!python/object/apply:src.misc.io.CustomPath |
| - checkpoints |
| - rebuttal |
| - dl3dv-8views |
| - submitted |
| use_plugins: false |
| log_slurm_id: true |
|
|
