【工程实践】Anomalib异常检测从训练到部署

OpenCV学堂

发布于 2026-04-02 20:07:26

1980

文章被收录于专栏：贾志刚-OpenCV学堂贾志刚-OpenCV学堂

Anomalib异常检测库介绍

异常检测库最新anomalib2.0 已经支持多种已经检测模型从训练到部署，同时支持多种推理方式。模型训练支持包含：

模型训练

以Padim模型为例，训练该模型支持CLI命令行与代码API两种方式，分别如下：

anomalib train --model Padim --data anomalib.data.MVTecAD

代码调用方式如下：

from anomalib.data import MVTecAD
from anomalib.models import Patchcore
from anomalib.engine import Engine
# Initialize components
datamodule = MVTecAD()
model = Patchcore()
engine = Engine()
# Train the model
engine.fit(datamodule=datamodule, model=model)

运行结果如下：

导出与部署

anomalib模型训练保存的是pytorch light格式的ckpt文件，支持导出torch、onnx、openvino三种格式。可以通过相关的SDK直接完成导出，以导出torch格式模型为例，导出CLI如下：

anomalib export --model Padim --export_type torch --ckpt_path <PATH_TO_CHECKPOINT>

导出以后模型模型推理支持部署到OpenVINO、Pytorch、TensorRT等，以pt文件与xml文件为例，推理演示代码如下：

# Get the inferencer.
inferencer = OpenVINOInferencer(path=args.weights, device=args.device)
filenames = get_image_filenames(path=args.input)
for filename in tqdm(filenames, desc="Predicting images"):
	predictions = inferencer.predict(filename)
# NOTE: This visualization approach is experimental and might change in the future.
	output = visualize_image_item(
		item=predictions.items[0],
		fields=["image"],  # Can be used to visualize other fields such as anomaly_map, pred_mask, etc.
		overlay_fields=[
# Can be used to overlay multiple other fields.
			("image", ["anomaly_map"]),
			("image", ["pred_mask"]),
		],
	)
if args.output is None and args.show is False:
		msg = "Neither output path is provided nor show flag is set. Inferencer will run but return nothing."
		logger.warning(msg)
if output is not None:
if args.output:
			file_path = generate_output_image_filename(input_path=filename, output_path=args.output)
			output.save(file_path)
# Show the image in case the flag is set by the user.
if args.show:
			output.show("Output Image")

效果如下：