configs/mvdiffusion-joint-ortho-6views.yaml

pretrained_model_name_or_path: 'lambdalabs/sd-image-variations-diffusers'
pretrained_unet_path: './ckpts/'
revision: null
validation_dataset:
  root_dir: "./example_images" # the folder path stores testing images
  num_views: 6
  bg_color: 'white'
  img_wh: [256, 256]
  num_validation_samples: 1000
  crop_size: 192
  filepaths: ['owl.png']  # the test image names. leave it empty, test all images in the folder

save_dir: 'outputs/'

pred_type: 'joint'
seed: 42
validation_batch_size: 1
dataloader_num_workers: 64

local_rank: -1

pipe_kwargs:
  camera_embedding_type: 'e_de_da_sincos'
  num_views: 6

validation_guidance_scales: [3.0]
pipe_validation_kwargs:
  eta: 1.0
validation_grid_nrow: 6

unet_from_pretrained_kwargs:
  camera_embedding_type: 'e_de_da_sincos'
  projection_class_embeddings_input_dim: 10  
  num_views: 6
  sample_size: 32
  zero_init_conv_in: false
  zero_init_camera_projection: false  

num_views: 6
camera_embedding_type: 'e_de_da_sincos'

enable_xformers_memory_efficient_attention: true