MCPcopy Create free account
hub / github.com/DSL-Lab/StreamSplat / Options

Class Options

configs/options.py:13–88  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

11
12@dataclass
13class Options:
14 model_type = "Encoder"
15 mixed_precision: str = 'bf16'
16 gradient_accumulation_steps: int = 1
17
18 # dataset
19 root_path: str = ""
20 re10k_path: str = ""
21 co3d_path: str = ""
22 davis_path: str = ""
23 vos_path: str = ""
24
25 batch_size: int = 32
26 num_workers: int = 8
27 resume: Optional[str] = None
28
29 # Data augmentation control
30 use_augmentation: bool = True
31
32 lr: float = 5e-4
33 enable_depth: bool = True
34 depth_loss_type: str = "ssitrim"
35 depth_model_name: str = 'vitl'
36 num_epochs: int = 250
37 warmup_iters: int = 50000
38 lr_decay_epochs: int = 250
39 gradient_clip: float = 1.0
40 orth_proj: bool = True
41 lambda_dssim: float = 0.2
42 forder: int = 1
43 dynamic_type: str = "poly"
44 image_height: int = 288
45 image_width: int = 512
46 down_resolution: Tuple[int, int] = (288, 512)
47 input_frames: int = 1
48 output_frames: int = 1
49 background_color: Tuple[float, float, float] = (0.0, 0.0, 0.0)
50 lambda_lpips: float = 0.05
51 lambda_depth: float = 0.05
52 lambda_reg: float = 0.0
53 ignore_large_loss: float = 0.0
54 lpips_start_epoch : int = 50
55 depth_start_epoch : int = 0
56 dino_name: str = 'dinov2_vitb14_reg'
57 workspace: str = './workspace'
58
59 # GS Encoder
60 in_channels: int = 3
61 patch_size: int = 8
62 checkpointing: bool = True
63 num_layers: int = 10
64 hidden_dim: int = 768
65 bwindow_size: Optional[int] = None # encoder window size
66 window_size: int = 2304 # upsampler window size
67 decoder_ratio: float = 2.0
68
69 # GS Decoder
70 decoder_dim: int = 64

Callers 1

options.pyFile · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected