About parsing output

Z.Y_W · October 7, 2025, 2:06am

When analyzing the detection task of YOLO11, I have two questions regarding the tflite format (with INT8 input and output):

For the INT8 output in the format of [batch, features, bbox], how can I determine the specific composition and order of the “features”?
After obtaining the output and performing dequantization, is it still necessary to normalize the results? (The reason is that after parsing the dequantization results, I found that some of the classification scores are greater than 1. I assume that the “features” consist of x, y, w, h, and all classification scores.)

Toxite · October 7, 2025, 5:18am

You can find an example with preprocessing and postprocessing here:

github.com/ultralytics/ultralytics

examples/YOLOv8-TFLite-Python/main.py

main

# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

from __future__ import annotations

import argparse

import cv2
import numpy as np
import yaml

from ultralytics.utils import ASSETS

try:
    from tflite_runtime.interpreter import Interpreter
except ImportError:
    import tensorflow as tf

    Interpreter = tf.lite.Interpreter

This file has been truncated. show original

Topic		Replies	Views
Exporting to tflite with int8 quantization for edge deployment Support yolo , question , support , troubleshooting	1	74	October 2, 2025
Difference in Keypoint Coordinate Formats Between YOLOv8 and YOLO11 TFLite Models Support question	8	304	November 21, 2024
Quantization YOLO yolo	11	1092	October 30, 2024
Feature vector shape in TensorRT converted pose estimation model? Support yolo , pose , code	4	103	April 1, 2025
Output of the model in training vs. inference YOLO question , curious	1	476	August 7, 2024

About parsing output

Related topics