File size: 3,849 Bytes
10ed3aa | 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 | """Small v1.1 reader; no model, GPU, raw data, or media decoding required."""
import json
from pathlib import Path, PurePosixPath
import torch
PROFILE = "pt/tensor_bundle/v1"
LAYOUT = {
"intrinsic": ("camera/camera.pt", "intrinsic"),
"keypoints_xyz": ("annotation/keypoints/keypoints.pt", "xyz"),
"mano": ("annotation/mano/mano.pt", "motion"),
}
def safe_path(root, reference):
"""Resolve a canonical, view-relative POSIX reference without traversal."""
if not isinstance(reference, str) or not reference or any(x in reference for x in ("\\", ":", "<", ">")):
raise ValueError(f"Invalid relative reference: {reference!r}")
path = PurePosixPath(reference)
if path.is_absolute() or ".." in path.parts or path.as_posix() != reference or reference == ".":
raise ValueError(f"Reference must stay inside view: {reference}")
result = Path(root).joinpath(*path.parts)
if not result.resolve().is_relative_to(Path(root).resolve()):
raise ValueError(f"Reference escapes its input tree: {reference}")
return result
def exact_tensor(a, b):
"""Compare dtype, shape and contiguous bytes, including NaN payload bits."""
return (a.dtype == b.dtype and a.shape == b.shape and
torch.equal(a.contiguous().view(torch.uint8), b.contiguous().view(torch.uint8)))
class ViewReader:
def __init__(self, view):
self.view = Path(view)
self.meta = json.loads((self.view / "metainfo.json").read_text())
if self.meta["schema_version"] != "hand-warehouse/1.1":
raise ValueError("ViewReader expects a v1.1 view")
self.cache = {}
def tensor(self, reference, keys=None):
path = safe_path(self.view, reference)
if keys is not None and (not isinstance(keys, list) or not keys or
any(not isinstance(k, str) or not k for k in keys)):
raise ValueError("Dictionary keys must be a nonempty list of nonempty strings")
if reference not in self.cache:
self.cache[reference] = torch.load(path, map_location="cpu", weights_only=True)
value = self.cache[reference]
for key in keys or []:
value = value[key]
if not isinstance(value, torch.Tensor) or value.device.type != "cpu":
raise TypeError(f"Reference is not a CPU tensor: {reference} {keys}")
return value
def numeric(self, name):
if name not in LAYOUT:
raise ValueError(f"Unsupported FreiHAND numerical component: {name}")
component = self.meta["components"][name]
reference, entry = LAYOUT[name]
if component["data_ref"] != reference or component["valid_ref"] != reference:
raise ValueError("Numerical data and valid must use the registered bundle path")
if component["data_key"] != [entry, "data"] or component["valid_key"] != [entry, "valid"]:
raise ValueError("Numerical data and valid must use the registered same-entry keys")
for key in ("storage", "valid_storage"):
if component[key]["profile"] != PROFILE:
raise ValueError("Numerical fields must declare the bundle profile")
data = self.tensor(component["data_ref"], component["data_key"])
valid = self.tensor(component["valid_ref"], component["valid_key"])
bundle = self.cache[reference]
if type(bundle) is not dict or set(bundle) != {entry} or type(bundle[entry]) is not dict or set(bundle[entry]) != {"data", "valid"}:
raise ValueError("Unexpected bundle fields")
if valid.dtype != torch.bool:
raise TypeError("valid must be bool")
return data, valid
def validity(self, name):
component = self.meta["components"][name]
return self.tensor(component["valid_ref"], component.get("valid_key"))
|