❓ How to modify num_workers for prediction?

[Cua1103]

Your Question

I am trying to use the following code for inference:
`if args.predict:
print(f"\n--- [阶段 B] 正在为 '{subfolder_name}' 加载模型并进行预测评估 ---")

            # 如果没有在本次运行中训练，则查找已存在的模型
            if not model_path:
                model_path = find_model_path(training_results_dir, subfolder_name)
            
            if not model_path:
                print(f"[错误] 无法为 '{subfolder_name}' 找到已训练的模型。跳过预测阶段。")
                continue

            # 如果没有在本次运行中训练，需要初始化模型对象
            if model is None:
                model = Patchcore(visualizer=False)

            print(f"用于预测的模型路径: {model_path}")

            normal_dir = Path(subfolder_path) / "test" / "good"
            abnormal_dir = Path(subfolder_path) / "test" / "bad"
            output_dir = Path(prediction_output_dir) / subfolder_name

            if not normal_dir.is_dir() or not abnormal_dir.is_dir():
                print(f"[警告] 测试目录 'test/good' 或 'test/bad' 不存在。跳过预测。")
                continue

            normal_dataset = PredictDataset(path=normal_dir, image_size=(256, 256))
            abnormal_dataset = PredictDataset(path=abnormal_dir, image_size=(256, 256))
            
            prediction_engine = Engine()
            normal_predictions = prediction_engine.predict(model=model, dataset=normal_dataset, ckpt_path=model_path)
            abnormal_predictions = prediction_engine.predict(model=model, dataset=abnormal_dataset, ckpt_path=model_path)

            true_labels, predicted_labels = [], []
            fp_image_paths, fn_image_paths = [], []`

But there is a warning:
/opt/anaconda3/envs/anomalib/lib/python3.10/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the num_workers argumenttonum_workers=255in theDataLoader` to improve performance.
Predicting DataLoader 0: 3%|█▍ | 323/11067 [00:09<05:14, 34.22it/s]

And I also find that the speed of DataLoader is slow (34.22it/s). I am certain that I do not have any hardware resource limitations. Can I get Any help?

Forum Check

• I have searched the discussions forum for an answer to my question.

[github-actions[bot]]

Thanks for submitting this issue! It has been added to our triage queue. A maintainer will review it shortly.

[rajeshgangireddy]

Hi @Cua1103
When you pass a dataset to the engine.predict() method, anomalib creates a torch Dataloader for your dataset :
DataLoader(dataset, collate_fn=dataset.collate_fn, pin_memory=True).
This means, the num_workers is the set to 0, as it's the default for torch DataLoader.

To pass num_workers, you can create and pass a dataloader directly to the engine.predict()
For example :

from torch.utils.data import DataLoader
normal_dataset = PredictDataset(path=normal_dir, image_size=(256, 256))
normal_dataloader = DataLoader(dataset, collate_fn=dataset.collate_fn, pin_memory=True, num_workers=8)

prediction_engine = Engine()
normal_predictions = prediction_engine.predict(model=model, dataloaders=normal_dataloader, ckpt_path=model_path)

Let me know how it goes.