Bug encountered in training detectron2: the test set cannot be evaluated during training

Let me write it out front

Recently in use detectron2 When training the network , I want to see the test effect of the model on the test set during the training , I am here yaml The configuration file is set to 5000 Time Iteratively output the evaluation results of a test set , The setup code is as follows ：

cfg.TEST.EVAL_PERIOD = 5000 # Printout 

But when I trained to the th 5000 The evaluation results of the test set did not appear as I expected during the first iteration , It shows a warning , An error is as follows ：

No evaluator found. Use DefaultTrainer.test(evaluators=), or implement its build_evaluator method.

resolvent

This report is wrong （warning） The source is detectron2/engine/defaults.py Official original build_evaluator function

Official primitive build_evaluator function

    @classmethod
    def build_evaluator(cls, cfg, dataset_name):
        """ Returns: DatasetEvaluator or None It is not implemented by default. """
        raise NotImplementedError(
            """ If you want DefaultTrainer to automatically run evaluation, please implement `build_evaluator()` in subclasses (see train_net.py for example). Alternatively, you can call evaluation functions yourself (see Colab balloon tutorial for example). """
        )

github The one above is right for this issue The discussion of the ：
https://github.com/facebookresearch/detectron2/issues/951

Reference resources github The answer above ：
See examples https://github.com/facebookresearch/detectron2/blob/master/tools/train_net.py

The main directory of the official source code has a train_net.py file , There are also build_evaluator function ,github The answer above means to refer to train_net.py Inside build_evaluator The contents of the function ．

My method is simple and rough , I am too lazy to refer to , Put the original build_evaluator Function delete , And then put train_net.py Inside build_evaluator Function to copy and paste to the original location

The modified build_evaluator function （ Follow train_net.py Inside build_evaluator The function is the same ）

    @classmethod
    def build_evaluator(cls, cfg, dataset_name, output_folder=None):
        """ Create evaluator(s) for a given dataset. This uses the special metadata "evaluator_type" associated with each builtin dataset. For your own dataset, you can simply create an evaluator manually in your script and do not have to worry about the hacky if-else logic here. """
        if output_folder is None:
            output_folder = os.path.join(cfg.OUTPUT_DIR, "inference")
        evaluator_list = []
        evaluator_type = MetadataCatalog.get(dataset_name).evaluator_type
        if evaluator_type in ["sem_seg", "coco_panoptic_seg"]:
            evaluator_list.append(
                SemSegEvaluator(
                    dataset_name,
                    distributed=True,
                    num_classes=cfg.MODEL.SEM_SEG_HEAD.NUM_CLASSES,
                    ignore_label=cfg.MODEL.SEM_SEG_HEAD.IGNORE_VALUE,
                    output_dir=output_folder,
                )
            )
        if evaluator_type in ["coco", "coco_panoptic_seg"]:
            evaluator_list.append(COCOEvaluator(dataset_name, cfg, True, output_folder))
        if evaluator_type == "coco_panoptic_seg":
            evaluator_list.append(COCOPanopticEvaluator(dataset_name, output_folder))
        if evaluator_type == "cityscapes_instance":
            assert (
                torch.cuda.device_count() >= comm.get_rank()
            ), "CityscapesEvaluator currently do not work with multiple machines."
            return CityscapesInstanceEvaluator(dataset_name)
        if evaluator_type == "cityscapes_sem_seg":
            assert (
                torch.cuda.device_count() >= comm.get_rank()
            ), "CityscapesEvaluator currently do not work with multiple machines."
            return CityscapesSemSegEvaluator(dataset_name)
        elif evaluator_type == "pascal_voc":
            return PascalVOCDetectionEvaluator(dataset_name)
        elif evaluator_type == "lvis":
            return LVISEvaluator(dataset_name, cfg, True, output_folder)
        if len(evaluator_list) == 0:
            raise NotImplementedError(
                "no Evaluator for the dataset {} with the type {}".format(
                    dataset_name, evaluator_type
                )
            )
        elif len(evaluator_list) == 1:
            return evaluator_list[0]
        return DatasetEvaluators(evaluator_list)

Be careful ： Some of the called functions need import Come in

''' Remember to load functions ！！！'''
from detectron2.evaluation import (
    CityscapesInstanceEvaluator,
    CityscapesSemSegEvaluator,
    COCOEvaluator,
    COCOPanopticEvaluator,
    DatasetEvaluators,
    LVISEvaluator,
    PascalVOCDetectionEvaluator,
    SemSegEvaluator,
    verify_results,
)

After the modification is completed , The problems encountered are solved ！！！