diff --git a/degirum_tools/detection_eval.py b/degirum_tools/detection_eval.py
index c834a1b..00aa41d 100644
--- a/degirum_tools/detection_eval.py
+++ b/degirum_tools/detection_eval.py
@@ -7,6 +7,7 @@
 
 import json, os, degirum as dg, numpy as np
 from typing import List, Optional
+from collections.abc import Collection
 from pycocotools.coco import COCO
 from pycocotools.cocoeval import COCOeval
 from pycocotools.mask import encode
@@ -40,6 +41,8 @@ def __init__(self, model: dg.model.Model, **kwargs):
         self.classmap: Optional[dict] = None
         # path to save the predictions as a JSON file
         self.pred_path: Optional[str] = None
+        # sigma values for each keypoint.
+        self.kpt_sigmas: Optional[Collection] = None
 
         if model.output_postprocess_type not in [
             "Detection",
@@ -123,7 +126,7 @@ def evaluate(
             # pose keypoint map calculation
             if ObjectDetectionModelEvaluator._is_pose_model(jdict[0]):
                 kp_stats = ObjectDetectionModelEvaluator._evaluate_coco(
-                    anno, pred, mAP_type="keypoints", img_id_list=sorted_img_id_list
+                    anno, pred, mAP_type="keypoints", img_id_list=sorted_img_id_list, kpt_sigmas=self.kpt_sigmas
                 )
                 stats.append(kp_stats)
             # instance segmentation map calculation
@@ -210,7 +213,11 @@ def _save_results_coco_json(results, jdict, image_id, class_map=None):
 
     @staticmethod
     def _evaluate_coco(
-        anno: COCO, pred: COCO, mAP_type: str = "bbox", img_id_list: List[str] = []
+        anno: COCO,
+        pred: COCO,
+        mAP_type: str = "bbox",
+        img_id_list: List[str] = [],
+        kpt_sigmas: Optional[Collection] = None
     ):
         """Evaluation process based on the ground truth COCO object and the prediction object
 
@@ -224,6 +231,8 @@ def _evaluate_coco(
         eval_obj = COCOeval(anno, pred, mAP_type)
         if img_id_list:
             eval_obj.params.imgIds = [id for id in img_id_list]  # image IDs to evaluate
+        if mAP_type == "keypoints" and kpt_sigmas is not None:
+            eval_obj.params.kpt_oks_sigmas = np.array(kpt_sigmas)
         eval_obj.evaluate()
         eval_obj.accumulate()
         eval_obj.summarize()