Fix size mismatch if the prediction is empty

[ZhiliangWu]

Currently, if the model output empty predictions like (tensor([]), tensor([]), tensor([])), the evaluation script would output errors like

Traceback (most recent call last):
  File "eval_ssd.py", line 243, in <module>
    mean_ap, class_ap = eval.compute()
  File "eval_ssd.py", line 66, in compute
    results = torch.cat(results)
RuntimeError: Sizes of tensors must match except in dimension 0. Expected size 7 but got size 3 for tensor number 472 in the list.

This fix adds an empty box with probability 0 to fix such errors, which would IMO not affect the final evaluation results.