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>
This commit is contained in:
Shaun Zhou
co-authored by
Claude Opus 5
Claude Opus 5
parent
57446c2f10
commit
86c86b105c
+33
-37
@@ -66,10 +66,13 @@ def test_load_config_detects_cycle(tmp_path):
|
||||
|
||||
|
||||
def test_project_configs_are_loadable():
|
||||
cfg = load_config("configs/train/baseline.yaml")
|
||||
assert cfg["model"]["name"] == "defect_net"
|
||||
assert cfg["data"]["train"]["dataset"]["name"] == "film_defect"
|
||||
assert len(cfg["classes"]) > 0
|
||||
cluster = load_config("configs/cluster/film.yaml")
|
||||
assert cluster["dataset"]["name"] == "unlabeled"
|
||||
assert cluster["num_clusters"] > 0
|
||||
|
||||
classify = load_config("configs/classify/film.yaml")
|
||||
assert classify["model"]["name"] == "classifier"
|
||||
assert classify["train"]["monitor"] == "macro_f1"
|
||||
|
||||
|
||||
def test_replace_key_swaps_a_whole_subdict(tmp_path):
|
||||
@@ -116,46 +119,39 @@ def test_registry_reports_incompatible_fields_actionably():
|
||||
def test_every_shipped_config_builds_its_components(config_path):
|
||||
"""仓库里每个配置都必须能真的构建出组件。
|
||||
|
||||
回归测试：configs/train/resnet18.yaml 曾经无法使用——它同时继承 tiny_conv 与
|
||||
resnet18 的模型配置，合并后 backbone 混了两边的字段，构建时直接报错。
|
||||
回归测试：派生配置曾经无法使用——它同时继承两份骨架配置，
|
||||
合并后 backbone 混了两边的字段（tiny_conv 的 widths 混进 torchvision），构建时直接报错。
|
||||
"""
|
||||
torch = pytest.importorskip("torch")
|
||||
from datasets.yolo import load_classes
|
||||
from models import build_model
|
||||
pytest.importorskip("torch")
|
||||
from models.registry import BACKBONES
|
||||
|
||||
cfg = load_config(config_path)
|
||||
|
||||
if "model" in cfg:
|
||||
if "classes" not in cfg:
|
||||
# configs/model/*.yaml 是片段，只能被 _base_ 组合，单独用没有 classes
|
||||
pytest.skip(f"{config_path} 是配置片段，不含 classes")
|
||||
model_cfg = dict(cfg["model"])
|
||||
model_cfg.setdefault("num_classes", len(load_classes(cfg["classes"])))
|
||||
# 单测里不下载 ImageNet 权重
|
||||
# 聚类配置：pretrained 用顶层 backbone，simclr 用 simclr.backbone
|
||||
for backbone_cfg in (cfg.get("backbone"), cfg.get_path("simclr.backbone")):
|
||||
if not backbone_cfg:
|
||||
continue
|
||||
backbone_cfg = dict(backbone_cfg)
|
||||
if backbone_cfg.get("name") == "torchvision":
|
||||
backbone_cfg["pretrained"] = False # 单测里不下载 ImageNet 权重
|
||||
backbone = BACKBONES.build(backbone_cfg)
|
||||
assert backbone.out_channels > 0
|
||||
|
||||
# 分类配置
|
||||
model_cfg = cfg.get("model")
|
||||
if model_cfg:
|
||||
import torch
|
||||
|
||||
from models import build_model
|
||||
|
||||
model_cfg = dict(model_cfg)
|
||||
model_cfg.setdefault("num_classes", 3)
|
||||
model_cfg["init_from"] = None
|
||||
if model_cfg.get("backbone", {}).get("name") == "torchvision":
|
||||
model_cfg["backbone"] = {**model_cfg["backbone"], "pretrained": False}
|
||||
model = build_model(model_cfg)
|
||||
with torch.no_grad():
|
||||
out = model(torch.randn(1, 3, 64, 64))
|
||||
assert "heatmap" in out
|
||||
|
||||
simclr_backbone = cfg.get_path("simclr.backbone")
|
||||
if simclr_backbone:
|
||||
backbone_cfg = dict(simclr_backbone)
|
||||
if backbone_cfg.get("name") == "torchvision":
|
||||
backbone_cfg["pretrained"] = False
|
||||
backbone = BACKBONES.build(backbone_cfg)
|
||||
assert backbone.out_channels > 0
|
||||
|
||||
|
||||
def test_eval_config_matches_its_training_backbone():
|
||||
"""评估配置构建出的骨架必须与训练配置一致，否则权重根本加载不上。"""
|
||||
for train_cfg, eval_cfg in [
|
||||
("configs/train/baseline.yaml", "configs/eval/baseline.yaml"),
|
||||
("configs/train/resnet18.yaml", "configs/eval/resnet18.yaml"),
|
||||
]:
|
||||
assert load_config(train_cfg)["model"] == load_config(eval_cfg)["model"], eval_cfg
|
||||
assert model(torch.randn(1, 3, 64, 64)).shape[1] == model_cfg["num_classes"]
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
@@ -169,13 +165,13 @@ def test_eval_config_matches_its_training_backbone():
|
||||
def test_set_rejects_typos_with_suggestion(override, expected_hint):
|
||||
"""--set 打错字母会凭空造出没人读的键，训练照跑但改动无效——必须拦下并给出建议。"""
|
||||
with pytest.raises(KeyError) as excinfo:
|
||||
load_config("configs/cluster/film_simclr_resnet50.yaml", [override])
|
||||
load_config("configs/cluster/film_simclr_resnet18.yaml", [override])
|
||||
assert expected_hint in str(excinfo.value)
|
||||
|
||||
|
||||
def test_set_accepts_existing_keys():
|
||||
cfg = load_config(
|
||||
"configs/cluster/film_simclr_resnet50.yaml",
|
||||
"configs/cluster/film_simclr_resnet18.yaml",
|
||||
["simclr.optimizer.lr=0.001", "simclr.optimizer.weight_decay=0.0", "num_clusters=7"],
|
||||
)
|
||||
assert cfg["simclr"]["optimizer"]["lr"] == 0.001
|
||||
|
||||
Reference in New Issue
Block a user