DreamCraft3D/configs/dreamcraft3d-texture.yaml

name: "dreamcraft3d-texture"
tag: "${rmspace:${system.prompt_processor.prompt},_}"
exp_root_dir: "outputs"
seed: 0

data_type: "single-image-datamodule"
data:
  image_path: ./load/images/hamburger_rgba.png
  height: 1024
  width: 1024
  default_elevation_deg: 0.0
  default_azimuth_deg: 0.0
  default_camera_distance: 3.8
  default_fovy_deg: 20.0
  requires_depth: false
  requires_normal: false
  use_mixed_camera_config: false
  random_camera:
    height: 1024
    width: 1024
    batch_size: 1
    eval_height: 1024
    eval_width: 1024
    eval_batch_size: 1
    elevation_range: [-10, 45]
    azimuth_range: [-180, 180]
    camera_distance_range: [3.8, 3.8]
    fovy_range: [20.0, 20.0] # Zero123 has fixed fovy
    progressive_until: 0
    camera_perturb: 0.0
    center_perturb: 0.0
    up_perturb: 0.0
    eval_elevation_deg: ${data.default_elevation_deg}
    eval_camera_distance: ${data.default_camera_distance}
    eval_fovy_deg: ${data.default_fovy_deg}
    batch_uniform_azimuth: false
    n_val_views: 40
    n_test_views: 120

system_type: "dreamcraft3d-system"
system:
  stage: texture
  use_mixed_camera_config: ${data.use_mixed_camera_config}
  geometry_convert_from: ???
  geometry_convert_inherit_texture: true
  geometry_type: "tetrahedra-sdf-grid"
  geometry:
    radius: 2.0 # consistent with coarse
    isosurface_resolution: 128
    isosurface_deformable_grid: true
    isosurface_remove_outliers: true
    pos_encoding_config:
      otype: HashGrid
      n_levels: 16
      n_features_per_level: 2
      log2_hashmap_size: 19
      base_resolution: 16
      per_level_scale: 1.447269237440378 # max resolution 4096
    fix_geometry: true

  material_type: "no-material"
  material:
    n_output_dims: 3

  background_type: "solid-color-background"

  renderer_type: "nvdiff-rasterizer"
  renderer:
    context_type: cuda

  prompt_processor_type: "stable-diffusion-prompt-processor"
  prompt_processor:
    pretrained_model_name_or_path: "stabilityai/stable-diffusion-2-1-base"
    prompt: ???
    front_threshold: 30.
    back_threshold: 30.

  guidance_type: "stable-diffusion-bsd-guidance"
  guidance:
    pretrained_model_name_or_path: "stabilityai/stable-diffusion-2-1-base"
    pretrained_model_name_or_path_lora: "stabilityai/stable-diffusion-2-1-base"
    # pretrained_model_name_or_path_lora: "stabilityai/stable-diffusion-2-1"
    guidance_scale: 2.0
    min_step_percent: 0.05
    max_step_percent: [0, 0.5, 0.2, 5000]
    only_pretrain_step: 1000

  # guidance_3d_type: "stable-zero123-guidance"
  # guidance_3d:
  #   pretrained_model_name_or_path: "./load/zero123/stable_zero123.ckpt"
  #   pretrained_config: "./load/zero123/sd-objaverse-finetune-c_concat-256.yaml"
  #   cond_image_path: ${data.image_path}
  #   cond_elevation_deg: ${data.default_elevation_deg}
  #   cond_azimuth_deg: ${data.default_azimuth_deg}
  #   cond_camera_distance: ${data.default_camera_distance}
  #   guidance_scale: 5.0
  #   min_step_percent: 0.2  # (start_iter, start_val, end_val, end_iter)
  #   max_step_percent: 0.5

  # control_guidance_type: "stable-diffusion-controlnet-reg-guidance"
  # control_guidance:
  #   min_step_percent: 0.1
  #   max_step_percent: 0.5
  # control_prompt_processor_type: "stable-diffusion-prompt-processor"
  # control_prompt_processor:
  #   pretrained_model_name_or_path: "SG161222/Realistic_Vision_V2.0"
  #   prompt: ${system.prompt_processor.prompt}
  #   front_threshold: 30.
  #   back_threshold: 30.

  freq:
    n_ref: 2
    ref_only_steps: 0
    ref_or_guidance: "alternate"
    no_diff_steps: -1
    guidance_eval: 0

  loggers:
    wandb:
      enable: false
      project: "threestudio"

  loss:
    lambda_sd: 0.01
    lambda_lora: 0.1
    lambda_pretrain: 0.1
    lambda_3d_sd: 0.0
    lambda_rgb: 1000.
    lambda_mask: 100.
    lambda_mask_binary: 0.0
    lambda_depth: 0.0
    lambda_depth_rel: 0.0
    lambda_normal: 0.0
    lambda_normal_smooth: 0.0
    lambda_3d_normal_smooth: 0.0
    lambda_z_variance: 0.0
    lambda_reg: 0.0

  optimizer:
    name: AdamW
    args:
      betas: [0.9, 0.99]
      eps: 1.e-4
    params:
      geometry.encoding:
        lr: 0.01
      geometry.feature_network:
        lr: 0.001
      guidance.train_unet:
        lr: 0.00001
      guidance.train_unet_lora:
        lr: 0.00001

trainer:
  max_steps: 5000
  log_every_n_steps: 1
  num_sanity_val_steps: 0
  val_check_interval: 200
  enable_progress_bar: true
  precision: 32
  strategy: "ddp_find_unused_parameters_true"

checkpoint:
  save_last: true
  save_top_k: -1
  every_n_train_steps: ${trainer.max_steps}