Error in prediction classes array for inference method in evaluator.py

## Issue with evaluation:

I am using detectron in order to make predictions for 1 class only (with a value of 0). When I try to use the `inference_on_dataset()` method, I am getting the following error:

```
[11/22 12:05:36 d2.evaluation.coco_evaluation]: Preparing results for COCO format ...
Validation run stopped due to:A prediction has class=62, but the dataset only has 1 classes and predicted class id should be in [0, 0].
```

When I try to look at outputs variable containing the pred_classes array in the `inference_on_dataset()` method ([line 165](https://github.com/facebookresearch/detectron2/blob/main/detectron2/evaluation/evaluator.py)), I see that it contains an array with some 0 values (correct class prediction), but some random values ranging from 0-100 as well, with the first number of the array being 62 (the number in the error). The number of zeros was shortened for readability:

```
tensor([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 62, 0, 60, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 59, 0, 0, 0, 0, 0, 0, 0, 0, 8, 0,
0, 0, 0, 60, 0, 53, 0, 0, 0, 25, 0, 60, 73, 58, 63, 0, 4, 0,
60, 69, 0, 0, 0, 6, 0, 37, 0, 0, 0, 0, 60, 74, 55, 60, 0, 0,
71, 0, 13, 0, 0, 57, 56, 2, 60, 73, 72, 13, 0, 60, 72, 39, 32, 37,
60, 37, 59, 13, 5, 50, 53, 73, 60, 60, 59])
```

The `pred_classes` array has the same size as the `pred_boxes` array, which shows it does contain the right predictions. When I replace the `pred_classes` array with an array of the same size filled with zeros (expected behavior), the method does not throw an error anymore.

How can I fix this issue?