Normalize top-1 metrics to [0, 1] (#1394)

* align metrics

* change check

* delete if

external/model-preparation-algorithm/mpa_tasks/extensions/datasets/mpa_cls_dataset.py

@@ -129,7 +129,11 @@ def evaluate(self, results, metric="accuracy", metric_options=None, logger=None)
             metrics.remove("class_accuracy")
             self.class_acc = True
 
+        # compute top-k metrics from mmcls and align them in [0,1] range
         eval_results = super().evaluate(results, metrics, metric_options, logger)
+        for k in metric_options['topk']:
+            eval_results[f'accuracy_top-{k}'] /= 100
+            assert 0 <= eval_results[f'accuracy_top-{k}'] <= 1
 
         # Add Evaluation Accuracy score per Class - it can be used only for multi-class dataset.
         if self.class_acc: