File size: 4,957 Bytes
59630ba | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 | experiment:
debug: false
tasks:
- training
num_nodes: 1
training:
precision: 16-mixed
compile: false
lr: 8.0e-06
batch_size: 2
max_epochs: 26
max_steps: -1
max_time: null
data:
num_workers: 11
shuffle: false
optim:
accumulate_grad_batches: 1
gradient_clip_val: 1.0
checkpointing:
every_n_train_steps: null
every_n_epochs: 1
train_time_interval: null
enable_version_counter: false
save_top_k: -1
validation:
precision: 16-mixed
compile: false
batch_size: 8
val_every_n_step: 1.0
val_every_n_epoch: 1
num_sanity_val_steps: 1
limit_batch: 1.0
data:
num_workers: 0
shuffle: false
inference_mode: true
test:
precision: 16-mixed
compile: false
batch_size: 4
limit_batch: 1.0
data:
num_workers: 0
shuffle: false
inference_mode: true
find_unused_parameters: false
reload_dataloaders_every_n_epochs: 1
ema:
enable: true
decay: 0.9999
validate_original_weights: false
dataset:
debug: false
save_dir: data/real-estate-10k
data_mean:
- - - 0.577
- - - 0.517
- - - 0.461
data_std:
- - - 0.249
- - - 0.249
- - - 0.268
latent:
enable: false
type: pre_sample
suffix: null
downsampling_factor:
- 1
- 8
num_channels: 4
resolution: 256
observation_shape:
- 3
- 256
- 256
max_frames: 16
n_frames: 400
context_length: 7
frame_skip: 1
filter_min_len: null
external_cond_dim: 16
external_cond_stack: false
external_cond_processing: null
preload: false
subdataset_size: 2400000
num_eval_videos: 100
maximize_training_data: true
augmentation:
frame_skip_increase: 1
horizontal_flip_prob: 0.5
reverse_prob: 0.5
back_and_forth_prob: 0.1
algorithm:
debug: false
lr: 8.0e-06
backbone:
name: u_vit3d_pose
channels:
- 128
- 256
- 576
- 1152
emb_channels: 1024
patch_size: 2
block_types:
- ResBlock
- ResBlock
- TransformerBlock
- TransformerBlock
block_dropouts:
- 0.0
- 0.0
- 0.1
- 0.1
num_updown_blocks:
- 3
- 3
- 6
num_mid_blocks: 20
num_heads: 9
pos_emb_type: rope
use_checkpointing:
- false
- false
- false
- false
conditioning:
dim: 180
external_cond_dropout: 0.1
use_fourier_noise_embedding: true
x_shape:
- 3
- 256
- 256
max_frames: 16
n_frames: 400
frame_skip: 1
context_frames: 7
latent:
enable: false
type: pre_sample
suffix: null
downsampling_factor:
- 1
- 8
num_channels: 4
data_mean:
- - - 0.577
- - - 0.517
- - - 0.461
data_std:
- - - 0.249
- - - 0.249
- - - 0.268
external_cond_dim: 16
external_cond_stack: false
external_cond_processing: null
compile: false
weight_decay: 0.01
optimizer_beta:
- 0.9
- 0.99
lr_scheduler:
name: constant_with_warmup
num_warmup_steps: 10000
num_training_steps: 550000
noise_level: random_independent
uniform_future:
enabled: false
fixed_context:
enabled: false
indices: null
dropout: 0
variable_context:
enabled: false
prob: 0
dropout: 0
chunk_size: -1
scheduling_matrix: full_sequence
replacement: noisy_scale
diffusion:
is_continuous: true
timesteps: 1000
beta_schedule: cosine_simple_diffusion
schedule_fn_kwargs:
shift: 1.0
shifted: 0.125
interpolated: false
use_causal_mask: false
clip_noise: 20.0
objective: pred_v
loss_weighting:
strategy: sigmoid
snr_clip: 5.0
cum_snr_decay: 0.9
sigmoid_bias: -1.0
sampling_timesteps: 50
ddim_sampling_eta: 0.0
reconstruction_guidance: 0.0
training_schedule:
name: cosine
shift: 0.125
precond_scale: 0.125
vae:
pretrained_path: null
pretrained_kwargs: {}
use_fp16: true
batch_size: 2
checkpoint:
reset_optimizer: false
strict: true
tasks:
prediction:
enabled: true
history_guidance:
name: conditional
keyframe_density: null
sliding_context_len: null
interpolation:
enabled: false
history_guidance:
name: conditional
max_batch_size: null
logging:
deterministic: 0
loss_freq: 100
grad_norm_freq: 100
max_num_videos: 256
n_metrics_frames: null
metrics: []
metrics_batch_size: 16
sanity_generation: false
raw_dir: null
camera_pose_conditioning:
normalize_by: first
bound: null
type: ray_encoding
alignment:
alignment_coeff: 0
encoder_type: vggt
apply_unnormalize_recon: true
alignment_context_length: 16
encoder_info:
- 24
- 512
- 512
mid_channels: 128
latents_info: null
_name: dfot_geometry_forcing
debug: false
wandb:
entity: replace_with_your_account
project: dfot
mode: offline
resume: null
load: null
|