Commit
·
296621f
1
Parent(s):
e5de8f8
Upload text2shape/car/cfg.yml with huggingface_hub
Browse files- text2shape/car/cfg.yml +285 -0
text2shape/car/cfg.yml
ADDED
@@ -0,0 +1,285 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
bash_name: ../exp/tmp/2022_0424_1457_15.sh
|
2 |
+
clipforge:
|
3 |
+
clip_model: ViT-B/32
|
4 |
+
enable: 1
|
5 |
+
feat_dim: 512
|
6 |
+
cmt: l2e-4
|
7 |
+
comet_key: ''
|
8 |
+
data:
|
9 |
+
batch_size: 10
|
10 |
+
batch_size_test: 10
|
11 |
+
cates: car
|
12 |
+
clip_forge_enable: 1
|
13 |
+
clip_model: ViT-B/32
|
14 |
+
cond_on_cat: 0
|
15 |
+
cond_on_voxel: 0
|
16 |
+
data_dir: data/ShapeNetCore.v2.PC15k
|
17 |
+
data_path: ''
|
18 |
+
dataset_scale: 1
|
19 |
+
dataset_type: shapenet15k
|
20 |
+
eval_test_split: 0
|
21 |
+
input_dim: 3
|
22 |
+
is_encode_whole_dataset_trainer: 0
|
23 |
+
is_load_pointflow2NS: 0
|
24 |
+
load_point_stat: ''
|
25 |
+
nclass: 55
|
26 |
+
noise_std: 0.1
|
27 |
+
noise_std_min: -1.0
|
28 |
+
noise_type: normal
|
29 |
+
normalize_global: true
|
30 |
+
normalize_per_shape: false
|
31 |
+
normalize_range: false
|
32 |
+
normalize_shape_box: false
|
33 |
+
normalize_std_per_axis: false
|
34 |
+
num_workers: 4
|
35 |
+
random_subsample: 1
|
36 |
+
recenter_per_shape: false
|
37 |
+
sample_with_replacement: 1
|
38 |
+
te_max_sample_points: 2048
|
39 |
+
tr_max_sample_points: 2048
|
40 |
+
train_drop_last: 1
|
41 |
+
type: datasets.neuralspline_datasets
|
42 |
+
voxel_size: 0.1
|
43 |
+
ddpm:
|
44 |
+
add_point_feat: true
|
45 |
+
attn:
|
46 |
+
- 0
|
47 |
+
- 1
|
48 |
+
- 0
|
49 |
+
- 0
|
50 |
+
beta_1: 0.0001
|
51 |
+
beta_T: 0.02
|
52 |
+
clip_denoised: 0
|
53 |
+
ddim_step: 200
|
54 |
+
dropout: 0.1
|
55 |
+
ema: 1
|
56 |
+
input_dim: 3
|
57 |
+
loss_type: l1_sum
|
58 |
+
loss_type_0: ''
|
59 |
+
loss_weight_cdnorm: 1.0
|
60 |
+
loss_weight_emd: 1.0
|
61 |
+
model_mean_type: eps
|
62 |
+
model_var_type: fixedlarge
|
63 |
+
ncenter:
|
64 |
+
- 1024
|
65 |
+
- 256
|
66 |
+
- 64
|
67 |
+
- 16
|
68 |
+
num_layers_classifier: 3
|
69 |
+
num_steps: 1000
|
70 |
+
p2_gamma: 1.0
|
71 |
+
p2_k: 1.0
|
72 |
+
pointnet_luo: 0
|
73 |
+
pointnet_plus: 1
|
74 |
+
pointnet_pp: 3
|
75 |
+
pvc:
|
76 |
+
mlp_after_pvc: 0
|
77 |
+
use_small_model: 0
|
78 |
+
sched_mode: linear
|
79 |
+
time_dim: 64
|
80 |
+
use_bn: false
|
81 |
+
use_global_attn: 0
|
82 |
+
use_gn: true
|
83 |
+
use_new_timeemb: 0
|
84 |
+
use_p2_weight: 0
|
85 |
+
use_pvc: 0
|
86 |
+
with_se: 0
|
87 |
+
dpm:
|
88 |
+
train_encoder_only: 0
|
89 |
+
dpm_ckpt: ''
|
90 |
+
eval:
|
91 |
+
load_other_vae_ckpt: 0
|
92 |
+
need_denoise: 0
|
93 |
+
eval_ddim_step: 0
|
94 |
+
eval_trainnll: 0
|
95 |
+
exp_name: 0424/nscar/751100h_train_l2e-4_vae_adainB20l1E3W8
|
96 |
+
has_shapelatent: 1
|
97 |
+
hash: 751100h
|
98 |
+
latent_pts:
|
99 |
+
ada_mlp_init_scale: 0.1
|
100 |
+
decoder_layer_out_dim: 32
|
101 |
+
dgcnn_last_hid: 0
|
102 |
+
encoder_layer_out_dim: 32
|
103 |
+
hid: 64
|
104 |
+
knn: 40
|
105 |
+
latent_as_pts: 0
|
106 |
+
latent_dim_ext:
|
107 |
+
- 64
|
108 |
+
mask_out_extra_latent: 0
|
109 |
+
n5layer: 1
|
110 |
+
normalization: bn
|
111 |
+
pts_sigma_offset: 0.0
|
112 |
+
pvd_mse_loss: 1
|
113 |
+
skip_weight: 0.01
|
114 |
+
style_dim: 128
|
115 |
+
style_encoder: models.shapelatent_modules.PointNetPlusEncoder
|
116 |
+
style_mlp: ''
|
117 |
+
style_prior: models.score_sde.resnet.PriorSEClip
|
118 |
+
use_linear_for_adagn: 0
|
119 |
+
weight_kl_feat: 1.0
|
120 |
+
weight_kl_glb: 1.0
|
121 |
+
weight_kl_pt: 1.0
|
122 |
+
log_dir: ../exp/0424/nscar/751100h_train_l2e-4_vae_adainB20l1E3W8
|
123 |
+
log_name: ../exp/0424/nscar/751100h_train_l2e-4_vae_adainB20l1E3W8
|
124 |
+
model_config: default
|
125 |
+
ngpu: 8
|
126 |
+
num_ref: 0
|
127 |
+
num_val_samples: 24
|
128 |
+
save_dir: ../exp/0424/nscar/751100h_train_l2e-4_vae_adainB20l1E3W8
|
129 |
+
sde:
|
130 |
+
attn_mhead: 0
|
131 |
+
attn_mhead_local: -1
|
132 |
+
autocast_train: false
|
133 |
+
beta_end: 20.0
|
134 |
+
beta_start: 0.1
|
135 |
+
bound_mlogit: 0
|
136 |
+
bound_mlogit_value: -5.42
|
137 |
+
condition_add: 1
|
138 |
+
condition_cat: 0
|
139 |
+
cont_kl_anneal: true
|
140 |
+
dae_checkpoint: ''
|
141 |
+
dataset: shape
|
142 |
+
ddim_kappa: 1.0
|
143 |
+
ddim_skip_type: uniform
|
144 |
+
denoising_stddevs: beta
|
145 |
+
diffusion_steps: 1000
|
146 |
+
drop_inactive_var: 0
|
147 |
+
dropout: 0.3
|
148 |
+
ema_decay: 0.9999
|
149 |
+
embedding_dim: 128
|
150 |
+
embedding_scale: 1.0
|
151 |
+
embedding_type: positional
|
152 |
+
epochs: 800
|
153 |
+
fir: false
|
154 |
+
global_prior_ckpt: ''
|
155 |
+
grad_clip_max_norm: 0.0
|
156 |
+
hier_prior: 0
|
157 |
+
hypara_mixing_logit: 0
|
158 |
+
init_t: 1.0
|
159 |
+
is_continues: 0
|
160 |
+
iw_sample_p: ll_iw
|
161 |
+
iw_sample_q: reweight_p_samples
|
162 |
+
iw_subvp_like_vp_sde: false
|
163 |
+
jac_reg_coeff: 0
|
164 |
+
jac_reg_freq: 1
|
165 |
+
kin_reg_coeff: 0
|
166 |
+
kl_anneal_portion_vada: 0.5
|
167 |
+
kl_balance_vada: false
|
168 |
+
kl_const_coeff_vada: 1.0e-07
|
169 |
+
kl_const_portion_vada: 0.0
|
170 |
+
kl_max_coeff_vada: 0.5
|
171 |
+
learn_mixing_logit: 1
|
172 |
+
learning_rate_dae: 0.0002
|
173 |
+
learning_rate_dae_local: 0.0003
|
174 |
+
learning_rate_min_dae: 0.0002
|
175 |
+
learning_rate_min_dae_local: 0.0003
|
176 |
+
learning_rate_min_vae: 1.0e-05
|
177 |
+
learning_rate_mlogit: -1.0
|
178 |
+
learning_rate_vae: 0.0001
|
179 |
+
local_prior: same_as_global
|
180 |
+
mixed_prediction: false
|
181 |
+
mixing_logit_init: -6
|
182 |
+
nhead: 4
|
183 |
+
num_cell_per_scale_dae: 8
|
184 |
+
num_cell_per_scale_dae_local: 0
|
185 |
+
num_channels_dae: 2048
|
186 |
+
num_latent_scales: 1
|
187 |
+
num_preprocess_blocks: 2
|
188 |
+
num_scales_dae: 2
|
189 |
+
ode_eps: 1.0e-05
|
190 |
+
ode_sample: 0
|
191 |
+
pool_feat_cat: 0
|
192 |
+
pos_embed: none
|
193 |
+
prior_model: models.latent_points_ada_localprior.PVCNN2Prior
|
194 |
+
progressive: none
|
195 |
+
progressive_combine: sum
|
196 |
+
progressive_input: none
|
197 |
+
regularize_mlogit: 0
|
198 |
+
regularize_mlogit_margin: 0.0
|
199 |
+
sde_type: vpsde
|
200 |
+
share_mlogit: 0
|
201 |
+
sigma2_0: 0.0
|
202 |
+
sigma2_max: 0.99
|
203 |
+
sigma2_min: 0.0001
|
204 |
+
time_emb_scales: 1.0
|
205 |
+
time_eps: 0.01
|
206 |
+
train_dae: 1
|
207 |
+
train_ode_solver_tol: 1.0e-05
|
208 |
+
train_vae: false
|
209 |
+
update_q_ema: false
|
210 |
+
use_adam: true
|
211 |
+
use_adamax: false
|
212 |
+
vae_checkpoint: ../exp/0415/nscar/4fec21h_hvae_kl0.5N32H1Anneall1_sumWlrInitScale_vae_adainB32l1E3W4/checkpoints/epoch_5999_iters_245999.pt
|
213 |
+
warmup_epochs: 20
|
214 |
+
weight_decay: 0.0003
|
215 |
+
weight_decay_norm_dae: 0.0
|
216 |
+
weight_decay_norm_vae: 0.0
|
217 |
+
set_detect_anomaly: 0
|
218 |
+
shapelatent:
|
219 |
+
decoder_num_points: 2048
|
220 |
+
decoder_type: models.latent_points_ada.LatentPointDecPVC
|
221 |
+
encoder_type: models.latent_points_ada.PointTransPVC
|
222 |
+
eps_z_global_only: 1
|
223 |
+
freeze_vae: 0
|
224 |
+
kl_weight: 0.5
|
225 |
+
latent_dim: 1
|
226 |
+
local_emb_agg: mean
|
227 |
+
log_sigma_offset: 6.0
|
228 |
+
loss0_weight: 1.0
|
229 |
+
model: models.vae_adain
|
230 |
+
prior_type: normal
|
231 |
+
residual: 1
|
232 |
+
snapshot_min: 30
|
233 |
+
test_size: 660
|
234 |
+
trainer:
|
235 |
+
anneal_kl: 1
|
236 |
+
apply_loss_weight_1_kl: 0
|
237 |
+
epochs: 18000
|
238 |
+
kl_balance: 0
|
239 |
+
kl_free:
|
240 |
+
- 0
|
241 |
+
- 0
|
242 |
+
kl_ratio:
|
243 |
+
- 1.0
|
244 |
+
- 1.0
|
245 |
+
kl_ratio_apply: 0
|
246 |
+
loss1_weight_anneal_v: quad
|
247 |
+
opt:
|
248 |
+
beta1: 0.9
|
249 |
+
beta2: 0.99
|
250 |
+
ema_decay: 0.9999
|
251 |
+
grad_clip: -1.0
|
252 |
+
lr: 0.001
|
253 |
+
lr_min: 0.0001
|
254 |
+
momentum: 0.9
|
255 |
+
scheduler: ''
|
256 |
+
start_ratio: 0.6
|
257 |
+
step_decay: 0.998
|
258 |
+
type: adam
|
259 |
+
vae_lr_warmup_epochs: 0
|
260 |
+
weight_decay: 0.0
|
261 |
+
rec_balance: 0
|
262 |
+
seed: 1
|
263 |
+
sn_reg_vae: 0
|
264 |
+
sn_reg_vae_weight: 0.0
|
265 |
+
type: trainers.train_2prior
|
266 |
+
use_grad_scalar: 0
|
267 |
+
use_kl_free: 0
|
268 |
+
warmup_epochs: 0
|
269 |
+
use_checkpoint: 0
|
270 |
+
vis_latent_point: 1
|
271 |
+
viz:
|
272 |
+
log_freq: -1
|
273 |
+
save_freq: 1000
|
274 |
+
val_freq: -10000
|
275 |
+
vis_sample_ddim_step: 0
|
276 |
+
viz_freq: -100
|
277 |
+
viz_order:
|
278 |
+
- 2
|
279 |
+
- 0
|
280 |
+
- 1
|
281 |
+
voxel2pts:
|
282 |
+
diffusion_steps:
|
283 |
+
- 0
|
284 |
+
init_weight: ''
|
285 |
+
weight_recont: 1.0
|