refactor: 按实际的五步工作流裁掉无关代码

实际工作流是：simclr 训练 -> 聚类 -> 人工归类 -> 分类训练 -> GradCAM/SHAP 生成
热力图与 box。据此移除不参与其中的检测/定位链路，共 31 个文件、约 2600 行：

- 模型：defect_net、postprocess、necks/、heads/center_head
- 数据：film_defect（YOLO 读取端）、target_encoder
- 引擎：trainer、evaluator、losses
- 入口：train.py、eval.py
- 脚本：convert_annotations、split_dataset、export_onnx、profile_inference
- 配置：configs/{train,eval,model,dataset}
- utils/labels（多类/单类标签模式，仅检测分类头使用）

共享模块按实际用量瘦身：engine/metrics 只留 AverageMeter（检测指标随之移除）；
datasets/yolo 只留写出侧（IMAGE_SUFFIXES 与 write_data_yaml）——第 5 步产出的
YOLO 标签是给外部检测训练用的交换格式，读取端不再需要。

裁剪过程中修掉一个真 bug：generate_boxes 的 --stage 默认 6（layer3）对 tiny_conv
这类非 nn.Sequential 骨架直接抛异常。现在自动退回骨架最终输出并警告，
另加 --stage -1 显式选择最终输出。

五步工作流已端到端实跑验证（simclr 1 轮 -> 聚类 -> 导出 -> 分类 1 轮 ->
gradcam 与 shap 两种方法各生成一遍）。

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tests/test_generate_boxes.py

@@ -94,21 +94,25 @@ def test_both_methods_produce_a_yolo_dataset(project, tmp_path, monkeypatch, met
             assert all(0.0 <= float(v) <= 1.0 for v in parts[1:])
 
 
-def test_output_is_readable_by_the_detection_dataset(project, tmp_path, monkeypatch):
-    """产出要能直接喂给检测训练——这是这一步存在的意义。"""
-    from datasets import FilmDefectDataset
+def test_output_is_a_wellformed_yolo_dataset(project, tmp_path, monkeypatch):
+    """产出是给外部检测训练用的交换格式，必须自洽：
+    data.yaml 的类别与 labels/ 对得上，每行标签可解析且坐标在 [0,1] 内。"""
+    import yaml
 
     out = tmp_path / "out"
     run(base_argv(project, out, "gradcam"), monkeypatch)
 
-    dataset = FilmDefectDataset(
-        img_dir=out / "images", classes=str(out / "data.yaml"),
-        transforms=[{"name": "Resize", "size": [64, 64]}, {"name": "ToTensor"}], stride=4,
-    )
-    assert len(dataset) == 5
-    sample = dataset[0]
-    assert sample["image"].shape == (3, 64, 64)
-    assert sample["boxes"].shape[1] == 4
+    meta = yaml.safe_load((out / "data.yaml").read_text(encoding="utf-8"))
+    names = [meta["names"][i] for i in sorted(meta["names"])]
+    assert meta["nc"] == len(names) == 2
+
+    for label in (out / "labels").rglob("*.txt"):
+        assert (out / "images" / label.parent.name / f"{label.stem}.png").is_file()
+        for line in label.read_text(encoding="utf-8").strip().splitlines():
+            class_id, cx, cy, w, h = line.split()
+            assert 0 <= int(class_id) < meta["nc"]
+            assert all(0.0 <= float(v) <= 1.0 for v in (cx, cy, w, h))
+            assert float(w) > 0 and float(h) > 0
 
 
 def test_images_are_hard_links_not_copies(project, tmp_path, monkeypatch):
@@ -224,3 +228,59 @@ def test_duplicate_source_names_are_rejected(tmp_path):
     gb._emit(source=Path("/data/dirA/x.png"), **common)
     with pytest.raises(gb.NameCollision, match="prefix-parent"):
         gb._emit(source=Path("/data/dirB/x.png"), **common)   # 同名，撞同一个标签路径
+
+
+def test_gradcam_falls_back_when_backbone_has_no_sequential_features(tmp_path, monkeypatch):
+    """tiny_conv 不是 nn.Sequential，取不了中间层——应当退回最终输出并警告，而不是崩掉。"""
+    from engine.checkpoint import save_checkpoint
+    from models import build_model
+
+    root = tmp_path / "dataset"
+    (root / "a").mkdir(parents=True)
+    rng = np.random.default_rng(0)
+    for i in range(2):
+        Image.fromarray(rng.integers(0, 255, (40, 56, 3), dtype=np.uint8)).save(
+            root / "a" / f"a_{i}.png"
+        )
+
+    backbone = {"name": "tiny_conv", "widths": [8, 16], "depths": [1, 1]}
+    checkpoint = tmp_path / "best.pth"
+    save_checkpoint(checkpoint, build_model(
+        {"name": "classifier", "num_classes": 1, "backbone": dict(backbone)}
+    ))
+
+    config = tmp_path / "cfg.yaml"
+    config.write_text(
+        f"""
+device: cpu
+data:
+  root: {root}
+  classes: null
+  exclude: []
+  val_transforms: [{{name: Resize, size: [64, 64]}}, {{name: ToTensor}}]
+  loader: {{batch_size: 2, num_workers: 0}}
+model:
+  name: classifier
+  backbone: {{name: tiny_conv, widths: [8, 16], depths: [1, 1]}}
+  init_from: null
+""",
+        encoding="utf-8",
+    )
+
+    out = tmp_path / "out"
+    monkeypatch.setattr(
+        sys, "argv",
+        ["generate_boxes.py", "--out", str(out), "--method", "gradcam",
+         "--checkpoint", str(checkpoint), "--config", str(config), "--no-heatmaps"],
+    )
+    assert gb.main() == 0                                    # 不崩
+
+    summary = json.loads((out / "summary.json").read_text("utf-8"))
+    assert summary["stage"] is None                          # 已退回最终输出
+    assert len(list((out / "labels").rglob("*.txt"))) == 2
+
+
+def test_stage_minus_one_selects_the_final_output(project, tmp_path, monkeypatch):
+    out = tmp_path / "out"
+    assert run(base_argv(project, out, "gradcam") + ["--stage", "-1"], monkeypatch) == 0
+    assert json.loads((out / "summary.json").read_text("utf-8"))["stage"] is None