File size: 1,667 Bytes
8b59682 |
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 |
CACHE_DIR=None
PRETRAINED_MODEL=None
num_frames = 16
frame_interval = 1
image_size = (256, 256)
fps = 30//2 # for inference
# Define dataset
root = None
data_path = "CSV_PATH"
use_image_transform = False
num_workers = 6
# Define acceleration
dtype = "fp16"
grad_checkpoint = False
plugin = "zero2"
sp_size = 1
data_prefetch = 1
MODEL_DIM = 1152
CAMERA_FORMAT = 'extrinsic'
CAMERA_PARAMS_NUM = 12
# Define model
model = dict(
type="STDiT-XL/2",
space_scale=0.5,
time_scale=1.0,
from_pretrained=PRETRAINED_MODEL,
enable_flashattn=True,
enable_layernorm_kernel=True,
camera_fuser_linear_dims=[MODEL_DIM+CAMERA_PARAMS_NUM, MODEL_DIM],
camera_format=CAMERA_FORMAT
)
vae = dict(
type="VideoAutoencoderKL",
from_pretrained="stabilityai/sd-vae-ft-ema",
cache_dir=CACHE_DIR,
)
text_encoder = dict(
type="t5",
from_pretrained="DeepFloyd/t5-v1_1-xxl",
model_max_length=120,
shardformer=True,
cache_dir=CACHE_DIR,
)
scheduler = dict(
type="iddpm_camera",
#num_sampling_steps=100,
cfg_scale_t=6.0,
cfg_scale_c=4.0
)
# Others
seed = 42
wandb = True
epochs = 12
log_every = 300
ckpt_every = 2000
dataset = dict(
text_dropout=0.05,
camera_dropout=0.05,
static_camera_rate=0.0,
resolution=256,
version='v0.7',
frame_strides=[4, 5, 6, 7, 8],
plucker_coord=False,
expand_rt=False
)
load = None
batch_size = 6
lr = 1e-5
grad_clip = 1.0
freeze_model = True
active_layer_names = ['camera_fuser', 'attn_temp']
# Inference
prompt_path = "./assets/texts/realestate10k.txt"
#prompt_path = "./assets/texts/t2v_sora.txt"
camera_path = ''
nprompts = None
save_dir = None |