conf/x/clipascene.yaml

method: 'clipascene'

im_name: ""
image_size: 224
u2net_path: "./checkpoint/u2net/u2net.pth"

background_layer: 2 # 2, 8, 11
background_div: 0.35 # 0.35, 0.5, 0.85
background_num_iter: 1501

foreground_layer: 2 # 2, 8, 11
foreground_div: 0.4 # 0.4, 0.5, 0.9
foreground_num_iter: 600 # 1000 if foreground_layer >= 8 else 600

# general
target: null
output_dir: null
path_svg: "none"
mask_object: 0
resize_obj: 0
fix_scale: 0
display_logs: 0
display: 0
test_name: "test"

# training
num_iter: 2001
num_stages: 1
lr_scheduler: 0
lr: 0.0001
color_lr: 0.01
width_lr: 0.0001
color_vars_threshold: 0.0
batch_size: 1
save_step: 100
eval_step: 20
loss_mask: "none"
dilated_mask: 0
mask_cls: None
mask_attention: 0

# strokes params
num_paths: 64
width: 1.5
control_points_per_seg: 4
num_segments: 1
attention_init: 1
saliency_model: "clip"
saliency_clip_model: "ViT-B/32"
xdog_intersec: 1
mask_object_attention: 0
softmax_temp: 0.3
mlp_train: 1
width_optim: 0
mlp_width_weights_path: "none"
mlp_points_weights_path: "none"
switch_loss: 0
gumbel_temp: 0.2
width_loss_weight: 0
width_loss_type: "L1"
optimize_points: 1
load_points_opt_weights: 0
gradnorm: 0
width_weights_lst: ""
ratio_loss: 0

# loss
percep_loss: "none"
perceptual_weight: 0
train_with_clip: 0
clip_weight: 0
start_clip: 0
num_aug_clip: 4
include_target_in_aug: 0
augment_both: 1
augemntations: "affine"
noise_thresh: 0.5
aug_scale_min: 0.7
force_sparse: 0
clip_conv_loss: 1
clip_mask_loss: 0
clip_conv_loss_type: "L2"
clip_conv_layer_weights: "0,0,1.0,1.0,0"
clip_model_name: "ViT-B/32"
clip_fc_loss_weight: 0
clip_text_guide: 0
text_target: None