How do I deal with situations where there is no object in the detected results?

[lartpang]

mmdetection/mmdet/core/evaluation/mean_ap.py

Line 308 in f240bf9

num_classes = len(det_results[0]) # positive class num

If there is no target in the detected result, an error will be reported here.

---

Can you give me some advice? Note that there is no object in the ground truth.

[ZwwWayne]

1. det_results means the detection results produced by models. If the results is produced by mmdetection models, det_results[0] will be a list of list, e.g., something like [[box for class 0],[box for class 1],[box for class 2],...,]. No matter whether there are boxes in the results, it can tell the number of classes.
2. Therefore, usually, you do not need do anything. Since you got an error here, something must go wrong. But we cannot do any help because the information is limited. Please use either Error Template or Reimplementation issues to report your error.

[lartpang]

@ZwwWayne Thanks for your reply.

In fact, I am using comments in pascal voc format.
Maybe there is something wrong with the way I generated the annotation file.
Can you give me some suggestions on how to construct a reasonable xml file for the data without a object?