rpx_benchmark.adapters.video_depth.skeletons
Skeleton adapter classes for every Video Depth model.
Same one-line-subclass pattern as the Image Depth skeletons; see
rpx_benchmark.adapters.depth.skeletons for the conventions
and rpx_benchmark.adapters.depth_scaffold for the base class.
1"""Skeleton adapter classes for every Video Depth model. 2 3Same one-line-subclass pattern as the Image Depth skeletons; see 4:mod:`rpx_benchmark.adapters.depth.skeletons` for the conventions 5and :mod:`rpx_benchmark.adapters.depth_scaffold` for the base class. 6""" 7 8from __future__ import annotations 9 10from ..depth_scaffold import DepthAdapterSkeleton 11 12 13class DA3VideoAdapter(DepthAdapterSkeleton): 14 """DA3 in video mode — same weights as ``da3-metric-l`` but fed a clip.""" 15 16 MODEL_KEY = "da3-video" 17 18 19class DepthCrafterAdapter(DepthAdapterSkeleton): 20 """DepthCrafter (Tencent, sliding-window diffusion, relative).""" 21 22 MODEL_KEY = "depth-crafter" 23 24 25class VideoDAAdapter(DepthAdapterSkeleton): 26 """Video Depth Anything.""" 27 28 MODEL_KEY = "video-da" 29 30 31class ChronoDepthAdapter(DepthAdapterSkeleton): 32 """ChronoDepth (sliding-window, relative).""" 33 34 MODEL_KEY = "chrono-depth" 35 36 37class RollingDepthAdapter(DepthAdapterSkeleton): 38 """RollingDepth (streaming, relative).""" 39 40 MODEL_KEY = "rolling-depth" 41 42 43class DVDAdapter(DepthAdapterSkeleton): 44 """DVD v1.1.""" 45 46 MODEL_KEY = "dvd" 47 48 49class GemDepthAdapter(DepthAdapterSkeleton): 50 """GemDepth.""" 51 52 MODEL_KEY = "gem-depth" 53 54 55class ViGeoAdapter(DepthAdapterSkeleton): 56 """ViGeo.""" 57 58 MODEL_KEY = "vigeo" 59 60 61class MonST3RAdapter(DepthAdapterSkeleton): 62 """MonST3R (multi-view 4-frame chunks).""" 63 64 MODEL_KEY = "monst3r" 65 66 67class VGGTOmegaAdapter(DepthAdapterSkeleton): 68 """VGGT-Ω (multi-view, large variant).""" 69 70 MODEL_KEY = "vggt-omega" 71 72 73__all__ = [ 74 "DA3VideoAdapter", 75 "DepthCrafterAdapter", 76 "VideoDAAdapter", 77 "ChronoDepthAdapter", 78 "RollingDepthAdapter", 79 "DVDAdapter", 80 "GemDepthAdapter", 81 "ViGeoAdapter", 82 "MonST3RAdapter", 83 "VGGTOmegaAdapter", 84]
14class DA3VideoAdapter(DepthAdapterSkeleton): 15 """DA3 in video mode — same weights as ``da3-metric-l`` but fed a clip.""" 16 17 MODEL_KEY = "da3-video"
DA3 in video mode — same weights as da3-metric-l but fed a clip.
install_hint: ClassVar[str] =
'pip install depth-anything-3; depth-anything/DA3-LARGE, fed per-clip instead of per-frame'
Inherited Members
20class DepthCrafterAdapter(DepthAdapterSkeleton): 21 """DepthCrafter (Tencent, sliding-window diffusion, relative).""" 22 23 MODEL_KEY = "depth-crafter"
DepthCrafter (Tencent, sliding-window diffusion, relative).
Inherited Members
26class VideoDAAdapter(DepthAdapterSkeleton): 27 """Video Depth Anything.""" 28 29 MODEL_KEY = "video-da"
Video Depth Anything.
Inherited Members
32class ChronoDepthAdapter(DepthAdapterSkeleton): 33 """ChronoDepth (sliding-window, relative).""" 34 35 MODEL_KEY = "chrono-depth"
ChronoDepth (sliding-window, relative).
Inherited Members
38class RollingDepthAdapter(DepthAdapterSkeleton): 39 """RollingDepth (streaming, relative).""" 40 41 MODEL_KEY = "rolling-depth"
RollingDepth (streaming, relative).
Inherited Members
DVD v1.1.
install_hint: ClassVar[str] =
'clone github.com/EnVision-Research/DVD at the pinned revision; official v1.1 weights: huggingface.co/FayeHongfeiZhang/DVD'
Inherited Members
GemDepth.
install_hint: ClassVar[str] =
'clone github.com/Yuecheng919/GemDepth; official weights: huggingface.co/YuechengLiu/GemDepth'
Inherited Members
ViGeo.
install_hint: ClassVar[str] =
'clone github.com/aigc3d/ViGeo; official weights: huggingface.co/pkqbajng/ViGeo'
Inherited Members
62class MonST3RAdapter(DepthAdapterSkeleton): 63 """MonST3R (multi-view 4-frame chunks).""" 64 65 MODEL_KEY = "monst3r"
MonST3R (multi-view 4-frame chunks).
install_hint: ClassVar[str] =
'pip install monst3r (or `pip install git+https://github.com/Junyi42/monst3r`)'
Inherited Members
68class VGGTOmegaAdapter(DepthAdapterSkeleton): 69 """VGGT-Ω (multi-view, large variant).""" 70 71 MODEL_KEY = "vggt-omega"
VGGT-Ω (multi-view, large variant).