Return empty iou when ground truth *or* detection is empty

Summary:
On horizontal box OCR datasets, d2go may fail in evaluation during the IoU calculation.

This happens when the detections are empty; the detections are empty when the image has been deleted, as in fbid [103133290522776](https://www.internalfb.com/intern/whatisthis/103133290522776), even though the the ground truth is non-empty. Therefore, in the case when `dt` OR `gt` is empty, we should return an empty list for the ious.

In addition, we should make sure the ious are type Tensor before calling `.cpu()`.

More context [here](https://fb.workplace.com/groups/277527419809135/posts/1376896089872257/?comment_id=1376905679871298&reply_comment_id=1377770513118148).

Reviewed By: Rechant

Differential Revision: D49613362

fbshipit-source-id: e2054f0ee28318a744e3259cc8949c00100d7b1f

detectron2/evaluation/rotated_coco_evaluation.py

@@ -65,16 +65,18 @@ def compute_iou_dt_gt(self, dt, gt, is_crowd):
             # This is the same as the classical COCO evaluation
             return maskUtils.iou(dt, gt, is_crowd)
 
-    def computeIoU(self, imgId, catId):
+    def computeIoU(self, imgId: int, catId: int):
         p = self.params
         if p.useCats:
             gt = self._gts[imgId, catId]
             dt = self._dts[imgId, catId]
         else:
             gt = [_ for cId in p.catIds for _ in self._gts[imgId, cId]]
             dt = [_ for cId in p.catIds for _ in self._dts[imgId, cId]]
-        if len(gt) == 0 and len(dt) == 0:
+
+        if len(gt) == 0 or len(dt) == 0:
             return []
+
         inds = np.argsort([-d["score"] for d in dt], kind="mergesort")
         dt = [dt[i] for i in inds]
         if len(dt) > p.maxDets[-1]: