iomgaa
1de88a5d0a
- 内存：prepare_segmentation 改为流式解码，只留分屏和配准的抽样帧，绿框逐帧在两半上找； 分割那一半由 SegmentationInput.target_frames() 到用时再解码，segment_video 边读边写 jpg； MedSAM2 init_state 的读帧换成直接在存放设备上建张量（不改 MedSAM2 源码，结果逐位相同）； 传播结果写进预分配数组。帧数最多的一段视频常驻内存峰值 11.2 GiB → 2.3 GiB，分辨率最大的 一段原代码被 OOM 杀掉、现为 4.2 GiB；两段的掩码和全部产物与改前逐字节相同，33 段视频的 第 1–2 步结果与改前一致，29 例配准与 review.csv 一致。 - 快照：ceus_quant 开跑时写 snapshot/（git、ceus.yaml、review.csv、uv.lock、segment.json）， 配置和核对表从快照副本读；收尾由新函数 runs.finish_snapshot 只新建 finished.json。 build 与 ceus_align_review 同样改用 finish_snapshot 记结束时间。 - build：summary.json 加 video_sha256；装 CEUS 结果时核对快照里的 delivery_dir、逐例核对 视频 SHA-256 和推注时间，不同就整次停下；新字段做类型检查，不进 tic.json。 来源：主会话任务（问题 1–3 及补充的 delivery_dir 比对）。 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
128 lines
5.5 KiB
Python
128 lines
5.5 KiB
Python
"""测试用假 CEUS 运行目录：布局照 `scripts/ceus_quant.py` 写出的（failed.csv、snapshot/、
|
||
`<视频号>/summary.json`、`<视频号>/keyframes/<槽位>.png`），数值全部编造，不调显卡。
|
||
|
||
summary.json 故意带上 tic.json 不该有的字段（视频号、视频 SHA-256、分割缓存键、帧号、拟合参数、
|
||
失败原因代码、推注时间、紫圈检出率），用来测 build 只按白名单挑字段。视频 SHA-256 和推注时间照
|
||
`fake_delivery` 的假交付包（视频内容 `VIDEO_BYTES`、推注 5 秒）。关键帧 PNG 带文本块和 EXIF。
|
||
"""
|
||
|
||
import csv
|
||
import hashlib
|
||
import json
|
||
from pathlib import Path
|
||
from typing import Any
|
||
|
||
from PIL import Image, PngImagePlugin
|
||
|
||
from fake_delivery import VIDEO_BYTES
|
||
|
||
SLOTS = ("B0", "D1", "D2", "D3", "D4", "F1", "F2")
|
||
CURVES = ("roi_all", "roi_perf", "roi_pan", "background_circle")
|
||
UNLOCATED_REASON = "拟合曲线上没有 D4 时刻（编造的原因）"
|
||
|
||
|
||
def _perfusion(scale: float) -> dict[str, Any]:
|
||
names = ("PE_au", "WiAUC_au_s", "RT_s", "mTTl_s", "TTP_s", "WiR_au_per_s", "WiPI_au")
|
||
more = ("WoAUC_au_s", "WiWoAUC_au_s", "FT_s", "WoR_au_per_s")
|
||
return {
|
||
"signal_domain": "db_display",
|
||
"time_origin": "video_start",
|
||
"area_curve": "fitted_minus_baseline",
|
||
"values": {name: scale * (i + 1) for i, name in enumerate((*names, *more))},
|
||
"supporting_times_s": {"TI": 3.0, "mTT": 9.5, "TO": None},
|
||
}
|
||
|
||
|
||
def _fit(success: bool, r2: float | None) -> dict[str, Any]:
|
||
return {
|
||
"success": success,
|
||
"failure_reason": None if success else "credible_peak_not_found",
|
||
"curve_model": "lognormal",
|
||
"baseline_source": "observed",
|
||
"valid_samples": 120,
|
||
"duplicate_samples": 3,
|
||
"parameters": {
|
||
"area": 1.0,
|
||
"mu": 2.0,
|
||
"sigma": 0.5,
|
||
"arrival_offset_s": 4.0,
|
||
"baseline": 9,
|
||
},
|
||
"quality": {"r2": r2, "nrmse": 0.1, "residual_noise_ratio": 1.2, "multi_peak_flag": False},
|
||
"model_arrival_time_video_s": 4.0,
|
||
"model_landmark_times_video_s": {"D2": 6.0, "D3": 9.0, "D4": 15.0},
|
||
"model_landmark_intensities": {"D2": 1.0, "D3": 2.0, "D4": 1.5},
|
||
}
|
||
|
||
|
||
def summary(video_id: str, located: tuple[str, ...]) -> dict[str, Any]:
|
||
"""一例的 summary.json：`located` 里的槽位定位到（视频时间=槽位序号×2.5 秒），余下未定位。"""
|
||
anchors = {
|
||
slot: {
|
||
"frame_index": index * 25 if slot in located else None,
|
||
"frame_time_video_s": index * 2.5 if slot in located else None,
|
||
"time_since_injection_s": index * 2.5 + 5.0 if slot in located else None,
|
||
"status": "ok" if slot in located else "anchor_undefined",
|
||
"reason": None if slot in located else UNLOCATED_REASON,
|
||
"anchor_time_video_s": index * 2.5,
|
||
"time_error_s": 0.0 if slot in located else None,
|
||
"source_curve": "roi_perf",
|
||
}
|
||
for index, slot in enumerate(SLOTS)
|
||
}
|
||
return {
|
||
"video_id": video_id,
|
||
"video_sha256": hashlib.sha256(VIDEO_BYTES).hexdigest(),
|
||
"segment_cache_key": "f" * 64,
|
||
"injection_offset_s": 5.0,
|
||
"reference_valid": False,
|
||
"pan_detected_fraction": 0.4,
|
||
"early_truncation_flag": False,
|
||
"lesion_fit_ready": True,
|
||
"fits": {
|
||
"roi_all": _fit(True, 0.93),
|
||
"roi_perf": _fit(True, 0.91),
|
||
"roi_pan": _fit(False, None),
|
||
"background_circle": _fit(False, 0.2),
|
||
},
|
||
"perfusion_metrics": {"roi_all": _perfusion(1.0), "roi_perf": None},
|
||
"anchor_frames": anchors,
|
||
"available_dynamic_slots": [s for s in ("B0", "D1", "D2", "D3", "D4") if s in located],
|
||
}
|
||
|
||
|
||
def keyframe_png(path: Path, shade: int) -> None:
|
||
"""一张 8 位 RGB 关键帧，带文本块和 EXIF（写进资料库前要去掉）。"""
|
||
pixels = bytes(v for i in range(12) for v in (shade, i * 20, 200 - i * 10))
|
||
image = Image.frombytes("RGB", (4, 3), pixels)
|
||
info = PngImagePlugin.PngInfo()
|
||
info.add_text("Source", "U999_roi2.mp4")
|
||
exif = Image.Exif()
|
||
exif[0x010E] = "U999" # ImageDescription
|
||
image.save(path, format="PNG", pnginfo=info, exif=exif.tobytes())
|
||
|
||
|
||
def write_run(
|
||
run_dir: Path,
|
||
delivery_dir: Path,
|
||
results: dict[str, tuple[str, ...]],
|
||
failed: list[tuple[str, int, str]],
|
||
) -> Path:
|
||
"""写一个已完成的运行目录：`delivery_dir` 记进快照里的 ceus.yaml，`results` 是视频号 → 定位到的
|
||
槽位，`failed` 是 failed.csv 的行。"""
|
||
(run_dir / "snapshot").mkdir(parents=True)
|
||
(run_dir / "snapshot" / "snapshot.json").write_text("{}", encoding="utf-8")
|
||
ceus_yaml = f"paths:\n delivery_dir: {json.dumps(str(delivery_dir), ensure_ascii=False)}\n"
|
||
(run_dir / "snapshot" / "ceus.yaml").write_text(ceus_yaml, encoding="utf-8")
|
||
with (run_dir / "failed.csv").open("w", encoding="utf-8", newline="") as handle:
|
||
writer = csv.writer(handle)
|
||
writer.writerow(("video_id", "step", "reason"))
|
||
writer.writerows(failed)
|
||
for video_id, located in results.items():
|
||
(run_dir / video_id / "keyframes").mkdir(parents=True)
|
||
content = json.dumps(summary(video_id, located), ensure_ascii=False, indent=2)
|
||
(run_dir / video_id / "summary.json").write_text(content, encoding="utf-8")
|
||
for shade, slot in enumerate(located):
|
||
keyframe_png(run_dir / video_id / "keyframes" / f"{slot}.png", shade * 30)
|
||
return run_dir
|