COCO format - annotation id

Consider the following object detection dataset in coco format (let's call it gt)

{"info": {}, "license": {},
"images": [{"id": 1, "width": 640, "height": 480, "file_name": "xyz.jpeg", "data_captured": null}],
"annotations": [{
    "id": 1, "image_id": 1, "category_id": 2,
    "bbox": [0.2, 0.2, 0.5, 0.2],
    "area": 0.01125, "iscrowd": 0
}],
"categories": [
    {"id": 1, "name": "category_1"},
    {"id": 2, "name": "category_2"}
]}

And predictions also in coco format

[{
    "image_id": 1, "category_id": 2,
    "bbox": [0.2, 0.2, 0.5, 0.2],
    "score": 0.99
}]

Evaluation on these files using pycocotools correctly outputs 100% mAP@IoU0.5 and other metrics.

If I replace gt['annotations'][0]['id'] to 0 like this

{"info": {}, "license": {},
"images": [{"id": 1, "width": 640, "height": 480, "file_name": "xyz.jpeg", "data_captured": null}],
"annotations": [{
    "id": 0, "image_id": 1, "category_id": 2,  # replace id here
    "bbox": [0.2, 0.2, 0.5, 0.2],
    "area": 0.01125, "iscrowd": 0
}],
"categories": [
    {"id": 1, "name": "category_1"},
    {"id": 2, "name": "category_2"}
]}

The evaluation returns 0 mAP@IoU0.5. Question: Is this bug in the pycocotools library or is there some special meaning to id=0 (similar to how category_id 0 is sometimes reserved for background?)

Minimal code sample to run an evaluation

!pip install pycocotools
import json
from pycocotools.cocoeval import COCOeval
from pycocotools.coco import COCO

gt_path = './gt.json'
labels_path = './labels.json'

coco_gt = COCO(gt_path)
anns = json.load(open(labels_path))
coco_dt = coco_gt.loadRes(labels_path)
coco_evaluator = COCOeval(coco_gt, coco_dt, 'bbox')
coco_evaluator.evaluate()
coco_evaluator.accumulate()
coco_evaluator.summarize()