kevin/operating-room-monitor-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

重命名 refs 为 algorithm_subprocesses，并清理误提交的权重与缓存文件。

Browse Source 
算法子进程目录仅保留源码与配置；权重、样本 I/O、构建产物通过 .gitignore 离线交付。

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Kevin
2026-05-21 16:10:08 +08:00
parent 1af442481e
commit 70431ca3f9
19 changed files with 88 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
backend/app/baked/algorithm.py
View File

@@ -60,7 +60,7 @@ HAND_DETECTION_MIN_CROP_PX: int = 64
HAND_DETECTION_DEVICE: str = ""
# --- refs/5.15 ActionFormer 实时版（VideoSwin → ActionFormer + 手检/好坏帧/耗材投票）---
# --- algorithm_subprocesses/5.15 ActionFormer 实时版（VideoSwin → ActionFormer + 手检/好坏帧/耗材投票）---
# 默认开启并与常见 demo 主摄 id 对齐；若缺少配置引用包 weights 下权重，开录时该路会记录异常。
ACTIONFORMER_ENABLED: bool = True
# 检查期：单路算法固定拉流 or-cam-03（恢复按请求首路时请改回 "" 或按需配置）。
@@ -69,7 +69,7 @@ ACTIONFORMER_PRIMARY_CAMERA_ID: str = "or-cam-03"
# RTSP 帧率假设（用于 frame_idx → t_abs 推算与 ActionFormer fps 入参；线上需与摄像机一致）。
ACTIONFORMER_ASSUMED_FPS: float = 25.0
# 与 refs/5.15 ``run_haocai_actionformer_consumables_e2e`` 写出的 infer_single.yaml 同语义的 VideoSwin 参数：
# 与 algorithm_subprocesses/5.15 ``run_haocai_actionformer_consumables_e2e`` 写出的 infer_single.yaml 同语义的 VideoSwin 参数：
# clip_len 32, frame_stride 2, feat_stride_frames 32 → feat_num_frames=64。
VIDEOSWIN_CLIP_LEN: int = 32
VIDEOSWIN_FRAME_STRIDE: int = 2
@@ -85,7 +85,7 @@ ACTIONFORMER_HAOCAI_WEIGHTS: str = default_actionformer_haocai_weights_path()
ACTIONFORMER_TEAR_WEIGHTS: str = default_actionformer_tear_weights_path()
ACTIONFORMER_NUM_CLASSES: int = 1 # 单类（Action）划段，与 e2e infer_single.yaml 对齐
# 与 refs/5.15 e2e 默认对齐的运行时阈值
# 与 algorithm_subprocesses/5.15 e2e 默认对齐的运行时阈值
ACTIONFORMER_DET_CONF: float = 0.5
ACTIONFORMER_PAD_RATIO: float = 0.30
ACTIONFORMER_DET_IMGSZ: int = 640