Fix mask dtype error with COCO validation.

coco.py

@@ -367,9 +367,11 @@ def evaluate_coco(model, dataset, coco, eval_type="bbox", limit=0, image_ids=Non
         t_prediction += (time.time() - t)
 
         # Convert results to COCO format
+        # Cast masks to uint8 because COCO tools errors out on bool
         image_results = build_coco_results(dataset, coco_image_ids[i:i + 1],
                                            r["rois"], r["class_ids"],
-                                           r["scores"], r["masks"])
+                                           r["scores"],
+                                           r["masks"].astype(np.uint8))
         results.extend(image_results)
 
     # Load results. This modifies results with additional attributes.