Detectron2 outputs validation loss during training

Let me write it out front

The problem lies in GitHub Of detectron2 Of issues Put forward on , Someone solved （ As shown in the figure below ）
Hint , Go to GitHub Upper issues Search for questions , Try to find 【closed】 Labeled , These are basically problems with solutions .
Just make a record here , For learning purposes only

Reference resources GitHub link ：
How do I compute validation loss during training?

This implementation is very clever , Directly replace the training set with the verification set , Use the calculation of the original training set loss To do the calculation

Added package

from detectron2.engine import HookBase
from detectron2.data import build_detection_train_loader
import detectron2.utils.comm as comm

Function function

class ValidationLoss(HookBase):
    def __init__(self, cfg, DATASETS_VAL_NAME):# Add one more DATASETS_VAL_NAME Parameters （ Small changes ）
        super().__init__()
        self.cfg = cfg.clone()
        self.cfg.DATASETS.TRAIN = DATASETS_VAL_NAME##
        self._loader = iter(build_detection_train_loader(self.cfg))
        
    def after_step(self):
        data = next(self._loader)
        with torch.no_grad():
            loss_dict = self.trainer.model(data)
            
            losses = sum(loss_dict.values())
            assert torch.isfinite(losses).all(), loss_dict

            loss_dict_reduced = {
    "val_" + k: v.item() for k, v in 
                                 comm.reduce_dict(loss_dict).items()}
            losses_reduced = sum(loss for loss in loss_dict_reduced.values())
            if comm.is_main_process():
                self.trainer.storage.put_scalars(total_val_loss=losses_reduced, 
                                                 **loss_dict_reduced)

Usage method

os.makedirs(cfg.OUTPUT_DIR, exist_ok=True)
trainer = Trainer(cfg) 
val_loss = ValidationLoss(cfg, cfg.DATASETS.VAL)  ## Additional parameters 
trainer.register_hooks([val_loss])
# swap the order of PeriodicWriter and ValidationLoss
trainer._hooks = trainer._hooks[:-2] + trainer._hooks[-2:][::-1]
trainer.resume_or_load(resume=False)
trainer.train()

Realization effect

total_val_loss
val_loss_cls
val_loss_box_reg