Reference for ultralytics/engine/validator.py

@smart_inference_mode()
def __call__(self, trainer=None, model=None):
    """Executes validation process, running inference on dataloader and computing performance metrics."""
    self.training = trainer is not None
    augment = self.args.augment and (not self.training)
    if self.training:
        self.device = trainer.device
        self.data = trainer.data
        # force FP16 val during training
        self.args.half = self.device.type != "cpu" and trainer.amp
        model = trainer.ema.ema or trainer.model
        model = model.half() if self.args.half else model.float()
        # self.model = model
        self.loss = torch.zeros_like(trainer.loss_items, device=trainer.device)
        self.args.plots &= trainer.stopper.possible_stop or (trainer.epoch == trainer.epochs - 1)
        model.eval()
    else:
        if str(self.args.model).endswith(".yaml") and model is None:
            LOGGER.warning("WARNING ⚠️ validating an untrained model YAML will result in 0 mAP.")
        callbacks.add_integration_callbacks(self)
        model = AutoBackend(
            weights=model or self.args.model,
            device=select_device(self.args.device, self.args.batch),
            dnn=self.args.dnn,
            data=self.args.data,
            fp16=self.args.half,
        )
        # self.model = model
        self.device = model.device  # update device
        self.args.half = model.fp16  # update half
        stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
        imgsz = check_imgsz(self.args.imgsz, stride=stride)
        if engine:
            self.args.batch = model.batch_size
        elif not pt and not jit:
            self.args.batch = model.metadata.get("batch", 1)  # export.py models default to batch-size 1
            LOGGER.info(f"Setting batch={self.args.batch} input of shape ({self.args.batch}, 3, {imgsz}, {imgsz})")

        if str(self.args.data).split(".")[-1] in {"yaml", "yml"}:
            self.data = check_det_dataset(self.args.data)
        elif self.args.task == "classify":
            self.data = check_cls_dataset(self.args.data, split=self.args.split)
        else:
            raise FileNotFoundError(emojis(f"Dataset '{self.args.data}' for task={self.args.task} not found ❌"))

        if self.device.type in {"cpu", "mps"}:
            self.args.workers = 0  # faster CPU val as time dominated by inference, not dataloading
        if not pt:
            self.args.rect = False
        self.stride = model.stride  # used in get_dataloader() for padding
        self.dataloader = self.dataloader or self.get_dataloader(self.data.get(self.args.split), self.args.batch)

        model.eval()
        model.warmup(imgsz=(1 if pt else self.args.batch, 3, imgsz, imgsz))  # warmup

    self.run_callbacks("on_val_start")
    dt = (
        Profile(device=self.device),
        Profile(device=self.device),
        Profile(device=self.device),
        Profile(device=self.device),
    )
    bar = TQDM(self.dataloader, desc=self.get_desc(), total=len(self.dataloader))
    self.init_metrics(de_parallel(model))
    self.jdict = []  # empty before each val
    for batch_i, batch in enumerate(bar):
        self.run_callbacks("on_val_batch_start")
        self.batch_i = batch_i
        # Preprocess
        with dt[0]:
            batch = self.preprocess(batch)

        # Inference
        with dt[1]:
            preds = model(batch["img"], augment=augment)

        # Loss
        with dt[2]:
            if self.training:
                self.loss += model.loss(batch, preds)[1]

        # Postprocess
        with dt[3]:
            preds = self.postprocess(preds)

        self.update_metrics(preds, batch)
        if self.args.plots and batch_i < 3:
            self.plot_val_samples(batch, batch_i)
            self.plot_predictions(batch, preds, batch_i)

        self.run_callbacks("on_val_batch_end")
    stats = self.get_stats()
    self.check_stats(stats)
    self.speed = dict(zip(self.speed.keys(), (x.t / len(self.dataloader.dataset) * 1e3 for x in dt)))
    self.finalize_metrics()
    self.print_results()
    self.run_callbacks("on_val_end")
    if self.training:
        model.float()
        results = {**stats, **trainer.label_loss_items(self.loss.cpu() / len(self.dataloader), prefix="val")}
        return {k: round(float(v), 5) for k, v in results.items()}  # return results as 5 decimal place floats
    else:
        LOGGER.info(
            "Speed: {:.1f}ms preprocess, {:.1f}ms inference, {:.1f}ms loss, {:.1f}ms postprocess per image".format(
                *tuple(self.speed.values())
            )
        )
        if self.args.save_json and self.jdict:
            with open(str(self.save_dir / "predictions.json"), "w") as f:
                LOGGER.info(f"Saving {f.name}...")
                json.dump(self.jdict, f)  # flatten and save
            stats = self.eval_json(stats)  # update stats
        if self.args.plots or self.args.save_json:
            LOGGER.info(f"Results saved to {colorstr('bold', self.save_dir)}")
        return stats

def add_callback(self, event: str, callback):
    """Appends the given callback."""
    self.callbacks[event].append(callback)

def build_dataset(self, img_path):
    """Build dataset."""
    raise NotImplementedError("build_dataset function not implemented in validator")

def check_stats(self, stats):
    """Checks statistics."""
    pass

def eval_json(self, stats):
    """Evaluate and return JSON format of prediction statistics."""
    pass

def finalize_metrics(self, *args, **kwargs):
    """Finalizes and returns all metrics."""
    pass

def get_dataloader(self, dataset_path, batch_size):
    """Get data loader from dataset path and batch size."""
    raise NotImplementedError("get_dataloader function not implemented for this validator")

def get_desc(self):
    """Get description of the YOLO model."""
    pass

def get_stats(self):
    """Returns statistics about the model's performance."""
    return {}

def init_metrics(self, model):
    """Initialize performance metrics for the YOLO model."""
    pass

def match_predictions(self, pred_classes, true_classes, iou, use_scipy=False):
    """
    Matches predictions to ground truth objects (pred_classes, true_classes) using IoU.

    Args:
        pred_classes (torch.Tensor): Predicted class indices of shape(N,).
        true_classes (torch.Tensor): Target class indices of shape(M,).
        iou (torch.Tensor): An NxM tensor containing the pairwise IoU values for predictions and ground of truth
        use_scipy (bool): Whether to use scipy for matching (more precise).

    Returns:
        (torch.Tensor): Correct tensor of shape(N,10) for 10 IoU thresholds.
    """
    # Dx10 matrix, where D - detections, 10 - IoU thresholds
    correct = np.zeros((pred_classes.shape[0], self.iouv.shape[0])).astype(bool)
    # LxD matrix where L - labels (rows), D - detections (columns)
    correct_class = true_classes[:, None] == pred_classes
    iou = iou * correct_class  # zero out the wrong classes
    iou = iou.cpu().numpy()
    for i, threshold in enumerate(self.iouv.cpu().tolist()):
        if use_scipy:
            # WARNING: known issue that reduces mAP in https://github.com/ultralytics/ultralytics/pull/4708
            import scipy  # scope import to avoid importing for all commands

            cost_matrix = iou * (iou >= threshold)
            if cost_matrix.any():
                labels_idx, detections_idx = scipy.optimize.linear_sum_assignment(cost_matrix, maximize=True)
                valid = cost_matrix[labels_idx, detections_idx] > 0
                if valid.any():
                    correct[detections_idx[valid], i] = True
        else:
            matches = np.nonzero(iou >= threshold)  # IoU > threshold and classes match
            matches = np.array(matches).T
            if matches.shape[0]:
                if matches.shape[0] > 1:
                    matches = matches[iou[matches[:, 0], matches[:, 1]].argsort()[::-1]]
                    matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
                    # matches = matches[matches[:, 2].argsort()[::-1]]
                    matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
                correct[matches[:, 1].astype(int), i] = True
    return torch.tensor(correct, dtype=torch.bool, device=pred_classes.device)

def on_plot(self, name, data=None):
    """Registers plots (e.g. to be consumed in callbacks)."""
    self.plots[Path(name)] = {"data": data, "timestamp": time.time()}

def plot_predictions(self, batch, preds, ni):
    """Plots YOLO model predictions on batch images."""
    pass

def plot_val_samples(self, batch, ni):
    """Plots validation samples during training."""
    pass

def postprocess(self, preds):
    """Preprocesses the predictions."""
    return preds

def pred_to_json(self, preds, batch):
    """Convert predictions to JSON format."""
    pass

def preprocess(self, batch):
    """Preprocesses an input batch."""
    return batch

def print_results(self):
    """Prints the results of the model's predictions."""
    pass

def run_callbacks(self, event: str):
    """Runs all callbacks associated with a specified event."""
    for callback in self.callbacks.get(event, []):
        callback(self)

def update_metrics(self, preds, batch):
    """Updates metrics based on predictions and batch."""
    pass