从Detectron2的COCOEvaluator获取输出指标

Question

标准输出的指标比写在输出文件夹中的输出有用得多。

Standard out

metrics.json和coco_instances_results.json

def test(instance, cfg, trainer, test_instance):  
  cfg.MODEL.WEIGHTS = os.path.join(cfg.OUTPUT_DIR, "model_final.pth")
  cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.7   # TODO What is this
  cfg.DATASETS.TEST = (test_instance, )
  predictor = DefaultPredictor(cfg)
  evaluator = COCOEvaluator(test_instance, cfg, False, output_dir=cfg.OUTPUT_DIR)
  val_loader = build_detection_test_loader(cfg, test_instance)
  result = inference_on_dataset(trainer.model, val_loader, evaluator)

如何获取ioU = 0.5的平均精度或按类别的bbox AP的详细信息？