Reference for ultralytics/models/yolo/pose/val.py

def _prepare_batch(self, si, batch):
    """Prepares a batch for processing by converting keypoints to float and moving to device."""
    pbatch = super()._prepare_batch(si, batch)
    kpts = batch["keypoints"][batch["batch_idx"] == si]
    h, w = pbatch["imgsz"]
    kpts = kpts.clone()
    kpts[..., 0] *= w
    kpts[..., 1] *= h
    kpts = ops.scale_coords(pbatch["imgsz"], kpts, pbatch["ori_shape"], ratio_pad=pbatch["ratio_pad"])
    pbatch["kpts"] = kpts
    return pbatch

def _prepare_pred(self, pred, pbatch):
    """Prepares and scales keypoints in a batch for pose processing."""
    predn = super()._prepare_pred(pred, pbatch)
    nk = pbatch["kpts"].shape[1]
    pred_kpts = predn[:, 6:].view(len(predn), nk, -1)
    ops.scale_coords(pbatch["imgsz"], pred_kpts, pbatch["ori_shape"], ratio_pad=pbatch["ratio_pad"])
    return predn, pred_kpts

def _process_batch(self, detections, gt_bboxes, gt_cls, pred_kpts=None, gt_kpts=None):
    """
    Return correct prediction matrix by computing Intersection over Union (IoU) between detections and ground truth.

    Args:
        detections (torch.Tensor): Tensor with shape (N, 6) representing detection boxes and scores, where each
            detection is of the format (x1, y1, x2, y2, conf, class).
        gt_bboxes (torch.Tensor): Tensor with shape (M, 4) representing ground truth bounding boxes, where each
            box is of the format (x1, y1, x2, y2).
        gt_cls (torch.Tensor): Tensor with shape (M,) representing ground truth class indices.
        pred_kpts (torch.Tensor | None): Optional tensor with shape (N, 51) representing predicted keypoints, where
            51 corresponds to 17 keypoints each having 3 values.
        gt_kpts (torch.Tensor | None): Optional tensor with shape (N, 51) representing ground truth keypoints.

    Returns:
        (torch.Tensor): A tensor with shape (N, 10) representing the correct prediction matrix for 10 IoU levels,
            where N is the number of detections.

    Example:
        ```python
        detections = torch.rand(100, 6)  # 100 predictions: (x1, y1, x2, y2, conf, class)
        gt_bboxes = torch.rand(50, 4)  # 50 ground truth boxes: (x1, y1, x2, y2)
        gt_cls = torch.randint(0, 2, (50,))  # 50 ground truth class indices
        pred_kpts = torch.rand(100, 51)  # 100 predicted keypoints
        gt_kpts = torch.rand(50, 51)  # 50 ground truth keypoints
        correct_preds = _process_batch(detections, gt_bboxes, gt_cls, pred_kpts, gt_kpts)
        ```

    Note:
        `0.53` scale factor used in area computation is referenced from https://github.com/jin-s13/xtcocoapi/blob/master/xtcocotools/cocoeval.py#L384.
    """
    if pred_kpts is not None and gt_kpts is not None:
        # `0.53` is from https://github.com/jin-s13/xtcocoapi/blob/master/xtcocotools/cocoeval.py#L384
        area = ops.xyxy2xywh(gt_bboxes)[:, 2:].prod(1) * 0.53
        iou = kpt_iou(gt_kpts, pred_kpts, sigma=self.sigma, area=area)
    else:  # boxes
        iou = box_iou(gt_bboxes, detections[:, :4])

    return self.match_predictions(detections[:, 5], gt_cls, iou)

def eval_json(self, stats):
    """Evaluates object detection model using COCO JSON format."""
    if self.args.save_json and self.is_coco and len(self.jdict):
        anno_json = self.data["path"] / "annotations/person_keypoints_val2017.json"  # annotations
        pred_json = self.save_dir / "predictions.json"  # predictions
        LOGGER.info(f"\nEvaluating pycocotools mAP using {pred_json} and {anno_json}...")
        try:  # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
            check_requirements("pycocotools>=2.0.6")
            from pycocotools.coco import COCO  # noqa
            from pycocotools.cocoeval import COCOeval  # noqa

            for x in anno_json, pred_json:
                assert x.is_file(), f"{x} file not found"
            anno = COCO(str(anno_json))  # init annotations api
            pred = anno.loadRes(str(pred_json))  # init predictions api (must pass string, not Path)
            for i, eval in enumerate([COCOeval(anno, pred, "bbox"), COCOeval(anno, pred, "keypoints")]):
                if self.is_coco:
                    eval.params.imgIds = [int(Path(x).stem) for x in self.dataloader.dataset.im_files]  # im to eval
                eval.evaluate()
                eval.accumulate()
                eval.summarize()
                idx = i * 4 + 2
                stats[self.metrics.keys[idx + 1]], stats[self.metrics.keys[idx]] = eval.stats[
                    :2
                ]  # update mAP50-95 and mAP50
        except Exception as e:
            LOGGER.warning(f"pycocotools unable to run: {e}")
    return stats

def get_desc(self):
    """Returns description of evaluation metrics in string format."""
    return ("%22s" + "%11s" * 10) % (
        "Class",
        "Images",
        "Instances",
        "Box(P",
        "R",
        "mAP50",
        "mAP50-95)",
        "Pose(P",
        "R",
        "mAP50",
        "mAP50-95)",
    )

def init_metrics(self, model):
    """Initiate pose estimation metrics for YOLO model."""
    super().init_metrics(model)
    self.kpt_shape = self.data["kpt_shape"]
    is_pose = self.kpt_shape == [17, 3]
    nkpt = self.kpt_shape[0]
    self.sigma = OKS_SIGMA if is_pose else np.ones(nkpt) / nkpt
    self.stats = dict(tp_p=[], tp=[], conf=[], pred_cls=[], target_cls=[], target_img=[])

def plot_predictions(self, batch, preds, ni):
    """Plots predictions for YOLO model."""
    pred_kpts = torch.cat([p[:, 6:].view(-1, *self.kpt_shape) for p in preds], 0)
    plot_images(
        batch["img"],
        *output_to_target(preds, max_det=self.args.max_det),
        kpts=pred_kpts,
        paths=batch["im_file"],
        fname=self.save_dir / f"val_batch{ni}_pred.jpg",
        names=self.names,
        on_plot=self.on_plot,
    )  # pred

def plot_val_samples(self, batch, ni):
    """Plots and saves validation set samples with predicted bounding boxes and keypoints."""
    plot_images(
        batch["img"],
        batch["batch_idx"],
        batch["cls"].squeeze(-1),
        batch["bboxes"],
        kpts=batch["keypoints"],
        paths=batch["im_file"],
        fname=self.save_dir / f"val_batch{ni}_labels.jpg",
        names=self.names,
        on_plot=self.on_plot,
    )

def pred_to_json(self, predn, filename):
    """Converts YOLO predictions to COCO JSON format."""
    stem = Path(filename).stem
    image_id = int(stem) if stem.isnumeric() else stem
    box = ops.xyxy2xywh(predn[:, :4])  # xywh
    box[:, :2] -= box[:, 2:] / 2  # xy center to top-left corner
    for p, b in zip(predn.tolist(), box.tolist()):
        self.jdict.append(
            {
                "image_id": image_id,
                "category_id": self.class_map[int(p[5])],
                "bbox": [round(x, 3) for x in b],
                "keypoints": p[6:],
                "score": round(p[4], 5),
            }
        )

def preprocess(self, batch):
    """Preprocesses the batch by converting the 'keypoints' data into a float and moving it to the device."""
    batch = super().preprocess(batch)
    batch["keypoints"] = batch["keypoints"].to(self.device).float()
    return batch

def save_one_txt(self, predn, pred_kpts, save_conf, shape, file):
    """Save YOLO detections to a txt file in normalized coordinates in a specific format."""
    from ultralytics.engine.results import Results

    Results(
        np.zeros((shape[0], shape[1]), dtype=np.uint8),
        path=None,
        names=self.names,
        boxes=predn[:, :6],
        keypoints=pred_kpts,
    ).save_txt(file, save_conf=save_conf)

def update_metrics(self, preds, batch):
    """Metrics."""
    for si, pred in enumerate(preds):
        self.seen += 1
        npr = len(pred)
        stat = dict(
            conf=torch.zeros(0, device=self.device),
            pred_cls=torch.zeros(0, device=self.device),
            tp=torch.zeros(npr, self.niou, dtype=torch.bool, device=self.device),
            tp_p=torch.zeros(npr, self.niou, dtype=torch.bool, device=self.device),
        )
        pbatch = self._prepare_batch(si, batch)
        cls, bbox = pbatch.pop("cls"), pbatch.pop("bbox")
        nl = len(cls)
        stat["target_cls"] = cls
        stat["target_img"] = cls.unique()
        if npr == 0:
            if nl:
                for k in self.stats.keys():
                    self.stats[k].append(stat[k])
                if self.args.plots:
                    self.confusion_matrix.process_batch(detections=None, gt_bboxes=bbox, gt_cls=cls)
            continue

        # Predictions
        if self.args.single_cls:
            pred[:, 5] = 0
        predn, pred_kpts = self._prepare_pred(pred, pbatch)
        stat["conf"] = predn[:, 4]
        stat["pred_cls"] = predn[:, 5]

        # Evaluate
        if nl:
            stat["tp"] = self._process_batch(predn, bbox, cls)
            stat["tp_p"] = self._process_batch(predn, bbox, cls, pred_kpts, pbatch["kpts"])
        if self.args.plots:
            self.confusion_matrix.process_batch(predn, bbox, cls)

        for k in self.stats.keys():
            self.stats[k].append(stat[k])

        # Save
        if self.args.save_json:
            self.pred_to_json(predn, batch["im_file"][si])
        if self.args.save_txt:
            self.save_one_txt(
                predn,
                pred_kpts,
                self.args.save_conf,
                pbatch["ori_shape"],
                self.save_dir / "labels" / f"{Path(batch['im_file'][si]).stem}.txt",
            )