5e2094f2ad
FastAPI service in infra/media-worker implementing every MediaInferenceProvider operation behind HTTP API v1, plus capabilities. Bearer auth, MEDIA_ROOT realpath enforcement, strict request schemas, cancellation on disconnect, GPU/CPU gates with 503 overflow, input size/duration limits, lazy SHA256-pinned model downloads, fail-closed redaction with atomic output, offline pytest suite with fake backends, CPU/CUDA/ROCm Dockerfile and a license review of every model. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0138YqDj3ttfayRwuMKKrqVj
55 lines
2.1 KiB
Python
55 lines
2.1 KiB
Python
"""Per-frame mask directories (`<tMs>.png`), as written by segmentPrompted."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import bisect
|
|
import os
|
|
import re
|
|
|
|
import cv2
|
|
import numpy as np
|
|
|
|
from ..errors import BadRequest, MediaError
|
|
|
|
MASK_NAME = re.compile(r"^(\d+)\.png$")
|
|
|
|
|
|
class MaskIndex:
|
|
def __init__(self, directory: str, frame_ms: float) -> None:
|
|
entries = []
|
|
for name in os.listdir(directory):
|
|
match = MASK_NAME.match(name)
|
|
if match:
|
|
entries.append((int(match.group(1)), os.path.join(directory, name)))
|
|
if not entries:
|
|
raise BadRequest("mask directory contains no <tMs>.png masks")
|
|
entries.sort()
|
|
self.times = [t for t, _ in entries]
|
|
self.files = [p for _, p in entries]
|
|
gaps = np.diff(self.times) if len(self.times) > 1 else np.array([frame_ms])
|
|
# A mask covers frames up to 1.5 sampling intervals away (at least one frame).
|
|
self.tolerance = max(float(np.median(gaps)) * 1.5, frame_ms)
|
|
self._cache: tuple[int, np.ndarray] | None = None
|
|
|
|
def lookup(self, t_ms: int, width: int, height: int) -> np.ndarray | None:
|
|
"""Binary mask (uint8 0/255) for the frame at t_ms, resized to the frame, or None."""
|
|
i = bisect.bisect_left(self.times, t_ms)
|
|
best = None
|
|
for j in (i - 1, i):
|
|
if not 0 <= j < len(self.times) or abs(self.times[j] - t_ms) > self.tolerance:
|
|
continue
|
|
if best is None or abs(self.times[j] - t_ms) < abs(self.times[best] - t_ms):
|
|
best = j
|
|
if best is None:
|
|
return None
|
|
if self._cache is not None and self._cache[0] == best:
|
|
return self._cache[1]
|
|
mask = cv2.imread(self.files[best], cv2.IMREAD_GRAYSCALE)
|
|
if mask is None:
|
|
raise MediaError("internal", f"unreadable mask {os.path.basename(self.files[best])}")
|
|
if mask.shape != (height, width):
|
|
mask = cv2.resize(mask, (width, height), interpolation=cv2.INTER_NEAREST)
|
|
mask = np.where(mask > 127, 255, 0).astype(np.uint8)
|
|
self._cache = (best, mask)
|
|
return mask
|