fix fish weight calculation by using correct parameters. output video

FishMeasure/README.md

@@ -1,7 +1,5 @@
 # 3D 鱼类重量测量系统
 
-> **仓库瘦身说明**：为便于部署到 `FishServer`，已删除历史推理目录（如 `output_weight_estimator/`、`output-yolo-sam/`）、训练数据集 `datasets/`、`project_jiuzhou01/` 前端工程及部分实验子项目数据。运行时仍需保留 **`sam_vit_h_4b8939.pth`**、`runs/train/.../best.pt`、`weight_estimator/runs/dgcnn_20260312_171043/best.pt` 等权重；详见仓库根目录 [`README.md`](../README.md)。
-
 基于 ZED 立体相机的 3D 鱼类重量自动测量项目。通过深度图检测、点云处理、模板匹配和体积计算，实现鱼类的非接触式重量估算。
 
 ## 项目概述

FishMeasure/detect_refbox/dataset/labels/train/fish10__HD1080_SN43186771_12-00-02__frame_002466.txt

@@ -0,0 +1 @@
+0 0.465612 0.844255 0.186298 0.259972

FishMeasure/detect_refbox/dataset/labels/train/fish10__HD1080_SN43186771_12-00-02__frame_002581.txt

@@ -0,0 +1 @@
+0 0.465278 0.841880 0.186966 0.259972

FishMeasure/detect_refbox/dataset/labels/train/fish10__HD1080_SN43186771_12-00-02__frame_002605.txt

@@ -0,0 +1 @@
+0 0.467949 0.842474 0.184295 0.256410

FishMeasure/detect_refbox/dataset/labels/train/fish10__HD1080_SN43186771_12-00-02__frame_004857.txt

@@ -0,0 +1 @@
+0 0.465278 0.842474 0.185630 0.254036

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-08-25__frame_000932.txt

@@ -0,0 +1 @@
+0 0.466279 0.838913 0.186298 0.256410

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-09-30__frame_005083.txt

@@ -0,0 +1 @@
+0 0.465612 0.841287 0.186298 0.256410

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-09-30__frame_005102.txt

@@ -0,0 +1 @@
+0 0.465278 0.844255 0.185630 0.257597

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-09-30__frame_005136.txt

@@ -0,0 +1 @@
+0 0.466279 0.841287 0.183627 0.254036

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-09-30__frame_005702.txt

@@ -0,0 +1 @@
+0 0.464276 0.840693 0.180956 0.252849

FishMeasure/detect_refbox/dataset/labels/train/fish11__HD1080_SN43186771_12-09-30__frame_005707.txt

@@ -0,0 +1 @@
+0 0.465946 0.843661 0.186966 0.254036

FishMeasure/detect_refbox/dataset/labels/train/fish16__HD1080_SN43186771_13-11-14__frame_003701.txt

@@ -0,0 +1 @@
+0 0.505709 0.935482 0.190371 0.129036

FishMeasure/detect_refbox/dataset/labels/train/fish16__HD1080_SN43186771_13-13-55__frame_000348.txt

@@ -0,0 +1 @@
+0 0.504340 0.935066 0.187634 0.121083

FishMeasure/detect_refbox/dataset/labels/train/fish16__HD1080_SN43186771_13-13-55__frame_000372.txt

@@ -0,0 +1 @@
+0 0.506010 0.935066 0.186966 0.123457

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-23-08__frame_003224.txt

@@ -0,0 +1 @@
+0 0.506273 0.937856 0.188827 0.124288

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-23-08__frame_003234.txt

@@ -0,0 +1 @@
+0 0.505330 0.937856 0.185607 0.124288

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-23-08__frame_003241.txt

@@ -0,0 +1 @@
+0 0.505342 0.934473 0.190972 0.127018

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-25-55__frame_001713.txt

@@ -0,0 +1 @@
+0 0.506010 0.934473 0.188301 0.115147

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-25-55__frame_001718.txt

@@ -0,0 +1 @@
+0 0.505008 0.936254 0.187634 0.116334

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-25-55__frame_001728.txt

@@ -0,0 +1 @@
+0 0.505398 0.937856 0.187079 0.124288

FishMeasure/detect_refbox/dataset/labels/train/fish17__HD1080_SN43186771_13-25-55__frame_001741.txt

@@ -0,0 +1 @@
+0 0.507032 0.934888 0.186339 0.130223

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-34-56__frame_000092.txt

@@ -0,0 +1 @@
+0 0.505283 0.939043 0.184177 0.121914

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-34-56__frame_000097.txt

@@ -0,0 +1 @@
+0 0.505008 0.936254 0.186298 0.118708

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-06__frame_000090.txt

@@ -0,0 +1 @@
+0 0.506677 0.937441 0.186966 0.121083

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-06__frame_000126.txt

@@ -0,0 +1 @@
+0 0.505342 0.936254 0.182959 0.125831

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000010.txt

@@ -0,0 +1 @@
+0 0.506415 0.937263 0.185105 0.125475

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000034.txt

@@ -0,0 +1 @@
+0 0.505676 0.935066 0.187634 0.125831

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000039.txt

@@ -0,0 +1 @@
+0 0.507011 0.936847 0.184963 0.124644

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000147.txt

@@ -0,0 +1 @@
+0 0.507011 0.936847 0.186298 0.124644

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000150.txt

@@ -0,0 +1 @@
+0 0.507413 0.937263 0.187103 0.125475

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000720.txt

@@ -0,0 +1 @@
+0 0.504930 0.936669 0.186142 0.126662

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000747.txt

@@ -0,0 +1 @@
+0 0.506010 0.936847 0.185630 0.117521

FishMeasure/detect_refbox/dataset/labels/train/fish18__HD1080_SN43186771_13-35-34__frame_000753.txt

@@ -0,0 +1 @@
+0 0.504830 0.937856 0.189949 0.124288

FishMeasure/detect_refbox/dataset/labels/train/fish20__HD1080_SN43186771_13-47-05__frame_001071.txt

@@ -0,0 +1 @@
+0 0.507180 0.937856 0.187971 0.124288

FishMeasure/detect_refbox/dataset/labels/train/fish8__HD1080_SN43186771_11-38-26__frame_001039.txt

@@ -0,0 +1 @@
+0 0.481862 0.954475 0.185412 0.091049

FishMeasure/detect_refbox/dataset/labels/train/fish8__HD1080_SN43186771_11-38-26__frame_001803.txt

@@ -0,0 +1 @@
+0 0.479464 0.953288 0.183287 0.093424

FishMeasure/detect_refbox/dataset/labels/train/fish8__HD1080_SN43186771_11-38-26__frame_001821.txt

@@ -0,0 +1 @@
+0 0.480728 0.953882 0.188486 0.092236

FishMeasure/detect_refbox/dataset/labels/train/fish8__HD1080_SN43186771_11-38-26__frame_001831.txt

@@ -0,0 +1 @@
+0 0.478966 0.954060 0.186298 0.090218

FishMeasure/detect_refbox/dataset/labels/val/fish10__HD1080_SN43186771_12-00-02__frame_002579.txt

@@ -0,0 +1 @@
+0 0.463942 0.843067 0.184295 0.255223

FishMeasure/detect_refbox/dataset/labels/val/fish11__HD1080_SN43186771_12-09-30__frame_001650.txt

@@ -0,0 +1 @@
+0 0.466613 0.843661 0.182959 0.261159

FishMeasure/detect_refbox/dataset/labels/val/fish11__HD1080_SN43186771_12-09-30__frame_004046.txt

@@ -0,0 +1 @@
+0 0.465946 0.839506 0.185630 0.255223

FishMeasure/detect_refbox/dataset/labels/val/fish11__HD1080_SN43186771_12-09-30__frame_005072.txt

@@ -0,0 +1 @@
+0 0.464610 0.843661 0.186966 0.256410

FishMeasure/detect_refbox/dataset/labels/val/fish16__HD1080_SN43186771_13-11-14__frame_002239.txt

@@ -0,0 +1 @@
+0 0.505756 0.936669 0.189129 0.126662

FishMeasure/detect_refbox/dataset/labels/val/fish16__HD1080_SN43186771_13-11-14__frame_002246.txt

@@ -0,0 +1 @@
+0 0.506010 0.935066 0.181624 0.123457

FishMeasure/detect_refbox/dataset/labels/val/fish16__HD1080_SN43186771_13-13-55__frame_000334.txt

@@ -0,0 +1 @@
+0 0.505452 0.936669 0.187186 0.126662

FishMeasure/detect_refbox/dataset/labels/val/fish18__HD1080_SN43186771_13-35-34__frame_000047.txt

@@ -0,0 +1 @@
+0 0.506484 0.936669 0.185243 0.126662

FishMeasure/detect_refbox/dataset/labels/val/fish19__HD1080_SN43186771_13-39-14__frame_001221.txt

@@ -0,0 +1 @@
+0 0.506677 0.933879 0.185630 0.116334

FishMeasure/detect_refbox/dataset/refbox.yaml

@@ -0,0 +1,6 @@
+path: /home/ubuntu/projects/FishMeasure/detect_refbox/dataset
+train: images/train
+val: images/val
+nc: 1
+names:
+  0: ref

FishMeasure/detect_refbox/prepare_refbox_dataset.py

@@ -0,0 +1,317 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+从递归目录中的 LabelMe 风格 JSON 构建 YOLO 检测数据集（仅参考物 ref）。
+
+- 仅使用同时满足：存在对应图像、JSON 内至少有一个可转换的 ref 矩形框 的样本。
+- 图像与 JSON 同目录，或通过 imagePath 解析；若仅有 imageData 则解码写出。
+- 输出扁平唯一文件名（相对路径转 __），避免不同子目录同名帧冲突。
+"""
+
+from __future__ import annotations
+
+import argparse
+import base64
+import io
+import json
+import random
+import shutil
+import sys
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Set, Tuple
+
+IMG_EXTS = {".jpg", ".jpeg", ".png", ".bmp", ".tif", ".tiff"}
+# 归一到单一类别 ref（class 0）
+REF_LABELS: Set[str] = {"ref", "reference", "refbox", "参考", "参考物"}
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description="准备 ref 框 YOLO 数据集")
+    p.add_argument(
+        "--source",
+        type=Path,
+        default=Path("/home/ubuntu/data/fish/2016-1-22-last_images"),
+        help="含递归子目录与 JSON 的根目录",
+    )
+    p.add_argument(
+        "--out",
+        type=Path,
+        default=None,
+        help="输出数据集根目录（默认：本仓库 detect_refbox/dataset）",
+    )
+    p.add_argument("--val-ratio", type=float, default=0.2, help="验证集比例")
+    p.add_argument("--seed", type=int, default=42, help="划分随机种子")
+    p.add_argument(
+        "--copy-images",
+        action="store_true",
+        help="复制图像；默认硬链接（同盘失败时回退复制）",
+    )
+    return p.parse_args()
+
+
+def repo_root() -> Path:
+    return Path(__file__).resolve().parents[1]
+
+
+def norm_bbox_yolo(
+    x1: float, y1: float, x2: float, y2: float, w: int, h: int
+) -> Tuple[float, float, float, float]:
+    x_min, y_min = min(x1, x2), min(y1, y2)
+    x_max, y_max = max(x1, x2), max(y1, y2)
+    bw = max(0.0, x_max - x_min)
+    bh = max(0.0, y_max - y_min)
+    cx = x_min + bw / 2.0
+    cy = y_min + bh / 2.0
+    if w <= 0 or h <= 0:
+        raise ValueError("invalid image size")
+    return cx / w, cy / h, bw / w, bh / h
+
+
+def load_json(path: Path) -> Optional[Dict[str, Any]]:
+    for enc in ("utf-8", "gbk", "gb2312", "latin-1"):
+        try:
+            with open(path, "r", encoding=enc) as f:
+                return json.load(f)
+        except (UnicodeDecodeError, json.JSONDecodeError):
+            continue
+    return None
+
+
+def find_image_path(json_path: Path, data: Dict[str, Any]) -> Optional[Path]:
+    ip = (data.get("imagePath") or "").strip()
+    if ip:
+        cand = (json_path.parent / ip).resolve()
+        if cand.exists() and cand.suffix.lower() in IMG_EXTS:
+            return cand
+        name = Path(ip).name
+        for ext in IMG_EXTS:
+            c2 = json_path.parent / f"{Path(name).stem}{ext}"
+            if c2.exists():
+                return c2
+    stem = json_path.stem
+    for ext in IMG_EXTS:
+        c = json_path.parent / f"{stem}{ext}"
+        if c.exists():
+            return c
+    return None
+
+
+def image_from_image_data(data: Dict[str, Any]) -> Optional[bytes]:
+    raw = data.get("imageData")
+    if not raw or not isinstance(raw, str):
+        return None
+    try:
+        return base64.b64decode(raw)
+    except Exception:
+        return None
+
+
+def shapes_to_yolo_lines(
+    data: Dict[str, Any], img_w: int, img_h: int
+) -> List[str]:
+    shapes = data.get("shapes") or data.get("annotations") or []
+    if not isinstance(shapes, list):
+        return []
+    lines: List[str] = []
+    for shp in shapes:
+        st = shp.get("shape_type")
+        if st not in (None, "rectangle", "bbox", "box"):
+            continue
+        label = str(shp.get("label", "")).strip()
+        if label not in REF_LABELS:
+            continue
+        pts = shp.get("points")
+        if not pts or len(pts) < 2:
+            continue
+        (x1, y1), (x2, y2) = pts[0], pts[1]
+        try:
+            xc, yc, bw, bh = norm_bbox_yolo(
+                float(x1), float(y1), float(x2), float(y2), img_w, img_h
+            )
+        except Exception:
+            continue
+        if bw <= 0 or bh <= 0:
+            continue
+        lines.append(f"0 {xc:.6f} {yc:.6f} {bw:.6f} {bh:.6f}")
+    return lines
+
+
+def ensure_image_size(
+    data: Dict[str, Any], img_path: Optional[Path]
+) -> Tuple[int, int]:
+    w = int(data.get("imageWidth", 0) or 0)
+    h = int(data.get("imageHeight", 0) or 0)
+    if w > 0 and h > 0:
+        return w, h
+    if img_path and img_path.exists():
+        try:
+            from PIL import Image
+
+            with Image.open(img_path) as im:
+                return im.size
+        except Exception:
+            pass
+    blob = image_from_image_data(data)
+    if blob:
+        try:
+            from PIL import Image
+
+            with Image.open(io.BytesIO(blob)) as im:
+                return im.size
+        except Exception:
+            pass
+    return 0, 0
+
+
+def unique_stem(json_path: Path, source_root: Path) -> str:
+    rel = json_path.parent.relative_to(source_root)
+    prefix = rel.as_posix().replace("/", "__")
+    return f"{prefix}__{json_path.stem}"
+
+
+def write_dataset_yaml(out_root: Path) -> Path:
+    yaml_path = out_root / "refbox.yaml"
+    # Ultralytics：path 为数据集根；train/val 为相对 path 的图像目录
+    text = (
+        f"path: {out_root.resolve()}\n"
+        "train: images/train\n"
+        "val: images/val\n"
+        "nc: 1\n"
+        "names:\n"
+        "  0: ref\n"
+    )
+    yaml_path.write_text(text, encoding="utf-8")
+    return yaml_path
+
+
+def main() -> int:
+    args = parse_args()
+    source = args.source.expanduser().resolve()
+    if not source.is_dir():
+        print(f"[错误] 数据目录不存在: {source}", file=sys.stderr)
+        return 1
+
+    out_root = (
+        args.out.expanduser().resolve()
+        if args.out
+        else repo_root() / "detect_refbox" / "dataset"
+    )
+    img_train = out_root / "images" / "train"
+    img_val = out_root / "images" / "val"
+    lbl_train = out_root / "labels" / "train"
+    lbl_val = out_root / "labels" / "val"
+    for d in (img_train, img_val, lbl_train, lbl_val):
+        d.mkdir(parents=True, exist_ok=True)
+
+    records: List[Tuple[Path, Path, List[str], str]] = []
+    # (json_path, src_image_path, yolo_lines, unique_stem)
+
+    json_files = sorted(source.rglob("*.json"))
+    skipped = 0
+    for jp in json_files:
+        data = load_json(jp)
+        if not data:
+            skipped += 1
+            continue
+
+        img_path = find_image_path(jp, data)
+        if not img_path:
+            blob = image_from_image_data(data)
+            if not blob:
+                skipped += 1
+                continue
+            ext = Path((data.get("imagePath") or "img.png")).suffix.lower()
+            if ext not in IMG_EXTS:
+                ext = ".png"
+            stem = unique_stem(jp, source)
+            tmp_img = out_root / "_tmp_decode" / f"{stem}{ext}"
+            tmp_img.parent.mkdir(parents=True, exist_ok=True)
+            tmp_img.write_bytes(blob)
+            img_path = tmp_img
+
+        iw, ih = ensure_image_size(data, img_path)
+        if iw <= 0 or ih <= 0:
+            skipped += 1
+            continue
+
+        lines = shapes_to_yolo_lines(data, iw, ih)
+        if not lines:
+            skipped += 1
+            continue
+
+        stem = unique_stem(jp, source)
+        ext = img_path.suffix.lower()
+        if ext not in IMG_EXTS:
+            ext = ".png"
+        records.append((jp, img_path, lines, stem + ext))
+
+    if not records:
+        print("[错误] 没有可用样本（需 JSON + 图像 + ref 矩形）", file=sys.stderr)
+        return 1
+
+    rng = random.Random(args.seed)
+    rng.shuffle(records)
+    n_val = int(round(len(records) * args.val_ratio))
+    n_val = max(1, n_val) if len(records) >= 2 else 0
+    n_val = min(n_val, len(records) - 1) if len(records) >= 2 else 0
+    # 仅 1 张：训练与验证共用同一张（写入两个目录），避免 YOLO 无 val
+    single_dup = len(records) == 1
+    val_set = set(range(len(records) - n_val, len(records))) if not single_dup else set()
+
+    n_tr = 0
+    n_va = 0
+
+    def materialize(
+        src_img: Path, lines: List[str], fname: str, is_val: bool
+    ) -> bool:
+        nonlocal n_tr, n_va, skipped
+        idir = img_val if is_val else img_train
+        ldir = lbl_val if is_val else lbl_train
+        dst_img = idir / fname
+        stem = Path(fname).stem
+        dst_lbl = ldir / f"{stem}.txt"
+        dst_img.parent.mkdir(parents=True, exist_ok=True)
+        try:
+            if args.copy_images:
+                shutil.copy2(src_img, dst_img)
+            else:
+                if dst_img.exists():
+                    dst_img.unlink()
+                try:
+                    dst_img.hardlink_to(src_img)
+                except OSError:
+                    shutil.copy2(src_img, dst_img)
+        except Exception as e:
+            print(f"[跳过] 复制图像失败 {src_img}: {e}", file=sys.stderr)
+            skipped += 1
+            return False
+        dst_lbl.write_text("\n".join(lines) + "\n", encoding="utf-8")
+        if is_val:
+            n_va += 1
+        else:
+            n_tr += 1
+        return True
+
+    for i, (_, src_img, lines, fname) in enumerate(records):
+        if single_dup:
+            materialize(src_img, lines, fname, is_val=False)
+            materialize(src_img, lines, fname, is_val=True)
+            break
+        is_val = i in val_set
+        materialize(src_img, lines, fname, is_val=is_val)
+
+    yaml_path = write_dataset_yaml(out_root)
+    tmp = out_root / "_tmp_decode"
+    if tmp.exists():
+        shutil.rmtree(tmp, ignore_errors=True)
+
+    print(
+        f"完成: 训练 {n_tr} / 验证 {n_va}（跳过 {skipped} 个 JSON）\n"
+        f"数据集: {out_root}\n"
+        f"YAML: {yaml_path}"
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())