configs/train_lora.yaml

dataset:
  num_views_low_res: 1
  num_views_high_res: 1
  image_root_dir: training/mp3d_skybox
  fov: 90
  rot_low_res: 45
  resolution: 256

train:
  log_dir: high_res_upsample
  lr: 0.0002

test:
  fuse_type: diffcollage # multidiff

model:
  model_id: stabilityai/stable-diffusion-2-base
  guidance_scale: 9.
  diff_timestep: 50
  model_type: base
  upsample_model:
    num_coarse_cp_blocks: 5
    lora_layers: True
    homo_cp_attn: True
  base_model:
    lora_layers: True
    single_image_ft: True