YOLOv8-16bit/ultralytics/yolo/v8/detect/val.py

import os
from pathlib import Path

import hydra
import numpy as np
import torch

from ultralytics.yolo.data import build_dataloader
from ultralytics.yolo.data.dataloaders.v5loader import create_dataloader
from ultralytics.yolo.engine.trainer import DEFAULT_CONFIG
from ultralytics.yolo.engine.validator import BaseValidator
from ultralytics.yolo.utils import colorstr, ops
from ultralytics.yolo.utils.checks import check_file, check_requirements
from ultralytics.yolo.utils.files import yaml_load
from ultralytics.yolo.utils.metrics import ConfusionMatrix, DetMetrics, box_iou
from ultralytics.yolo.utils.plotting import output_to_target, plot_images
from ultralytics.yolo.utils.torch_utils import de_parallel


class DetectionValidator(BaseValidator):

    def __init__(self, dataloader=None, save_dir=None, pbar=None, logger=None, args=None):
        super().__init__(dataloader, save_dir, pbar, logger, args)
        self.data_dict = yaml_load(check_file(self.args.data)) if self.args.data else None
        self.is_coco = False
        self.class_map = None
        self.targets = None
        self.metrics = DetMetrics(save_dir=self.save_dir, plot=self.args.plots)
        self.iouv = torch.linspace(0.5, 0.95, 10)  # iou vector for mAP@0.5:0.95
        self.niou = self.iouv.numel()

    def preprocess(self, batch):
        batch["img"] = batch["img"].to(self.device, non_blocking=True)
        batch["img"] = (batch["img"].half() if self.args.half else batch["img"].float()) / 255
        self.nb, _, self.height, self.width = batch["img"].shape  # batch size, channels, height, width
        self.targets = torch.cat((batch["batch_idx"].view(-1, 1), batch["cls"].view(-1, 1), batch["bboxes"]), 1)
        self.targets = self.targets.to(self.device)
        height, width = batch["img"].shape[2:]
        self.targets[:, 2:] *= torch.tensor((width, height, width, height), device=self.device)  # to pixels
        self.lb = [self.targets[self.targets[:, 0] == i, 1:]
                   for i in range(self.nb)] if self.args.save_hybrid else []  # for autolabelling

        return batch

    def init_metrics(self, model):
        head = model.model[-1] if self.training else model.model.model[-1]
        if self.data:
            self.is_coco = self.data.get('val', '').endswith(f'coco{os.sep}val2017.txt')  # is COCO dataset
            self.class_map = ops.coco80_to_coco91_class() if self.is_coco else list(range(1000))
            self.args.save_json |= self.is_coco and not self.training  # run on final val if training COCO
        self.nc = head.nc
        self.names = model.names
        self.metrics.names = self.names
        self.confusion_matrix = ConfusionMatrix(nc=self.nc)
        self.seen = 0
        self.jdict = []
        self.stats = []

    def get_desc(self):
        return ('%22s' + '%11s' * 6) % ('Class', 'Images', 'Instances', 'Box(P', "R", "mAP50", "mAP50-95)")

    def postprocess(self, preds):
        preds = ops.non_max_suppression(preds,
                                        self.args.conf_thres,
                                        self.args.iou_thres,
                                        labels=self.lb,
                                        multi_label=True,
                                        agnostic=self.args.single_cls,
                                        max_det=self.args.max_det)
        return preds

    def update_metrics(self, preds, batch):
        # Metrics
        for si, (pred) in enumerate(preds):
            labels = self.targets[self.targets[:, 0] == si, 1:]
            nl, npr = labels.shape[0], pred.shape[0]  # number of labels, predictions
            shape = batch["ori_shape"][si]
            # path = batch["shape"][si][0]
            correct_bboxes = torch.zeros(npr, self.niou, dtype=torch.bool, device=self.device)  # init
            self.seen += 1

            if npr == 0:
                if nl:
                    self.stats.append((correct_bboxes, *torch.zeros((2, 0), device=self.device), labels[:, 0]))
                    if self.args.plots:
                        self.confusion_matrix.process_batch(detections=None, labels=labels[:, 0])
                continue

            # Predictions
            if self.args.single_cls:
                pred[:, 5] = 0
            predn = pred.clone()
            ops.scale_boxes(batch["img"][si].shape[1:], predn[:, :4], shape)  # native-space pred

            # Evaluate
            if nl:
                tbox = ops.xywh2xyxy(labels[:, 1:5])  # target boxes
                ops.scale_boxes(batch["img"][si].shape[1:], tbox, shape)  # native-space labels
                labelsn = torch.cat((labels[:, 0:1], tbox), 1)  # native-space labels
                correct_bboxes = self._process_batch(predn, labelsn)
                # TODO: maybe remove these `self.` arguments as they already are member variable
                if self.args.plots:
                    self.confusion_matrix.process_batch(predn, labelsn)
            self.stats.append((correct_bboxes, pred[:, 4], pred[:, 5], labels[:, 0]))  # (conf, pcls, tcls)

            # TODO: Save/log
            '''
            if self.args.save_txt:
                save_one_txt(predn, save_conf, shape, file=save_dir / 'labels' / f'{path.stem}.txt')
            '''

    def get_stats(self):
        stats = [torch.cat(x, 0).cpu().numpy() for x in zip(*self.stats)]  # to numpy
        if len(stats) and stats[0].any():
            self.metrics.process(*stats)
        self.nt_per_class = np.bincount(stats[-1].astype(int), minlength=self.nc)  # number of targets per class
        fitness = {"fitness": self.metrics.fitness()}
        metrics = dict(zip(self.metric_keys, self.metrics.mean_results()))
        return {**metrics, **fitness}

    def print_results(self):
        pf = '%22s' + '%11i' * 2 + '%11.3g' * len(self.metric_keys)  # print format
        self.logger.info(pf % ("all", self.seen, self.nt_per_class.sum(), *self.metrics.mean_results()))
        if self.nt_per_class.sum() == 0:
            self.logger.warning(
                f'WARNING ⚠️ no labels found in {self.args.task} set, can not compute metrics without labels')

        # Print results per class
        if (self.args.verbose or not self.training) and self.nc > 1 and len(self.stats):
            for i, c in enumerate(self.metrics.ap_class_index):
                self.logger.info(pf % (self.names[c], self.seen, self.nt_per_class[c], *self.metrics.class_result(i)))

        if self.args.plots:
            self.confusion_matrix.plot(save_dir=self.save_dir, names=list(self.names.values()))

    def _process_batch(self, detections, labels):
        """
        Return correct prediction matrix
        Arguments:
            detections (array[N, 6]), x1, y1, x2, y2, conf, class
            labels (array[M, 5]), class, x1, y1, x2, y2
        Returns:
            correct (array[N, 10]), for 10 IoU levels
        """
        iou = box_iou(labels[:, 1:], detections[:, :4])
        correct = np.zeros((detections.shape[0], self.iouv.shape[0])).astype(bool)
        correct_class = labels[:, 0:1] == detections[:, 5]
        for i in range(len(self.iouv)):
            x = torch.where((iou >= self.iouv[i]) & correct_class)  # IoU > threshold and classes match
            if x[0].shape[0]:
                matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]),
                                    1).cpu().numpy()  # [label, detect, iou]
                if x[0].shape[0] > 1:
                    matches = matches[matches[:, 2].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=detections.device)

    def get_dataloader(self, dataset_path, batch_size):
        # TODO: manage splits differently
        # calculate stride - check if model is initialized
        gs = max(int(de_parallel(self.model).stride if self.model else 0), 32)
        return create_dataloader(path=dataset_path,
                                 imgsz=self.args.imgsz,
                                 batch_size=batch_size,
                                 stride=gs,
                                 hyp=dict(self.args),
                                 cache=False,
                                 pad=0.5,
                                 rect=self.args.rect,
                                 workers=self.args.workers,
                                 prefix=colorstr(f'{val}: '),
                                 shuffle=False,
                                 seed=self.args.seed)[0] if self.args.v5loader else \
            build_dataloader(self.args, batch_size, img_path=dataset_path, stride=gs, mode="val")[0]

    # TODO: align with train loss metrics
    @property
    def metric_keys(self):
        return ["metrics/precision(B)", "metrics/recall(B)", "metrics/mAP50(B)", "metrics/mAP50-95(B)"]

    def plot_val_samples(self, batch, ni):
        plot_images(batch["img"],
                    batch["batch_idx"],
                    batch["cls"].squeeze(-1),
                    batch["bboxes"],
                    paths=batch["im_file"],
                    fname=self.save_dir / f"val_batch{ni}_labels.jpg",
                    names=self.names)

    def plot_predictions(self, batch, preds, ni):
        plot_images(batch["img"],
                    *output_to_target(preds, max_det=15),
                    paths=batch["im_file"],
                    fname=self.save_dir / f'val_batch{ni}_pred.jpg',
                    names=self.names)  # pred

    def pred_to_json(self, preds, batch):
        for i, f in enumerate(batch["im_file"]):
            stem = Path(f).stem
            image_id = int(stem) if stem.isnumeric() else stem
            box = ops.xyxy2xywh(preds[i][:, :4])  # xywh
            box[:, :2] -= box[:, 2:] / 2  # xy center to top-left corner
            for p, b in zip(preds[i].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],
                    'score': round(p[4], 5)})

    def eval_json(self):
        if self.args.save_json and self.is_coco and len(self.jdict):
            anno_json = self.data['path'] / "annotations/instances_val2017.json"  # annotations
            pred_json = self.save_dir / "predictions.json"  # predictions
            self.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')
                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)
                eval = COCOeval(anno, pred, 'bbox')
                if self.is_coco:
                    eval.params.imgIds = [int(Path(x).stem) for x in self.dataloader.dataset.im_files]  # images to eval
                eval.evaluate()
                eval.accumulate()
                eval.summarize()
                self.metrics.metric.map, self.metrics.metric.map50 = eval.stats[:2]  # update mAP50-95 and mAP50
            except Exception as e:
                self.logger.warning(f'pycocotools unable to run: {e}')


@hydra.main(version_base=None, config_path=str(DEFAULT_CONFIG.parent), config_name=DEFAULT_CONFIG.name)
def val(cfg):
    cfg.data = cfg.data or "coco128.yaml"
    validator = DetectionValidator(args=cfg)
    validator(model=cfg.model)


if __name__ == "__main__":
    val()