替换SlowFast中Detectron2为Yolov8

一 需求

FaceBookReserch中SlowFast源码中检测框是用Detectron2进行目标检测，本文想实现用yolov8替换detectron2
1

二 实施方案

首先，yolov8 支持有自定义库ultralytics（仅支持yolov8），安装对应库

pip install ultralytics
1

源码中slowfast/visualization.py 43行中

if cfg.DETECTION.ENABLE:
       self.object_detector = Detectron2Predictor(cfg, gpu_id=self.gpu_id)
1
2

根据ultralytics文档进行定义
创建对应YOLOPredictor类(加入了检测框及其标签，具体见前一篇文章)

class YOLOPredictor:

    def __init__(self, cfg, gpu_id=None):
        # 加载预训练的 YOLOv8n 模型
        self.model = YOLO('/root/autodl-tmp/data/runs/detect/train/weights/best.pt')
        self.detect_names, _, _ = get_class_names(cfg.DEMO.Detect_File_Path, None, None)

    def __call__(self, task):
        """
        Return bounding boxes predictions as a tensor.
        Args:
            task (TaskInfo object): task object that contain
                the necessary information for action prediction. (e.g. frames)
        Returns:
            task (TaskInfo object): the same task info object but filled with
                prediction values (a tensor) and the corresponding boxes for
                action detection task.
        """
        # """得到预测置信度"""
        # scores = outputs["instances"].scores[mask].tolist()
        # """获取类别标签"""
        # pred_labels = outputs["instances"].pred_classes[mask]
        # pred_labels = pred_labels.tolist()
        # """进行标签匹配"""
        # for i in range(len(pred_labels)):
        #     pred_labels[i] = self.detect_names[pred_labels[i]]
        # preds = [
        #     "[{:.4f}] {}".format(s, labels) for s, labels in zip(scores, pred_labels)
        # ]
        # """加入预测标签"""
        # task.add_detect_preds(preds)
        # task.add_bboxes(pred_boxes)
        middle_frame = task.frames[len(task.frames) // 2]
        outputs = self.model(middle_frame)
        boxes = outputs[0].boxes
        mask = boxes.conf >= 0.5
        pred_boxes = boxes.xyxy[mask]
        scores = boxes.conf[mask].tolist()
        pred_labels = boxes.cls[mask].to(torch.int)
        pred_labels = pred_labels.tolist()
        for i in range(len(pred_labels)):
            pred_labels[i] = self.detect_names[pred_labels[i]]
        preds = [
            "[{:.4f}] {}".format(s, labels) for s, labels in zip(scores, pred_labels)
        ]
        """加入预测标签"""
        task.add_detect_preds(preds)
        task.add_bboxes(pred_boxes)

        return task
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50