多个app图像识别热量功能准确度分析数据

该数据可应用于携带图像识别能力的烹饪设备、手机或大模型、软件等软硬件制造开发企业。通过测试不同大模型对图片上食材热量的识别准确度，对比图片上食材的热量实际值和测试值的偏差，分析相应设备或者软件图像识别能力，可将这些数据喂至企业自身大模型，应用于图像识别能力的大模型或AI工具的能力训练。1、根据中国营养协会公布的食材热量表，计算不同实验的食物实际热量值Q1：Q1=m1/100*q（单位 kcal）；q是食物对应每百克热量（单位：kcal），查表可获得；m1是实验食物的实际重量m1。
2、将实验中食物进行拍照，进而将带有食物的图片发送给不同的大模型/agent，大模型/agent对于图片中食物种类、食物识别数量、食物识别重量m2进行识别判定，同时给出该图片上食物的测试测试值Q2。计算大模型/agent热量识别过程中的食物对应每百克热量q1（单位：kcal）：q1=100*Q2/m2。记录大模型/agent对图片中食物种类识别是否准确(输出的名称能基本吻合，如戚风蛋糕、海绵蛋糕能识别成蛋糕)。
3、计算大模型/agent对图片中食物重量识别准确率A：食物重量识别准确率A=m2-m1/m1；进而判断重量识别是否准确。若|A|≤50%，则该组数据对应的“重量识别是否准确结果”为“是”；若|A|>50%，则该组数据对应的“重量识别是否准确的结果”为“否”。

This dataset can be applied to software and hardware manufacturing and development enterprises that produce cooking appliances, mobile phones, large language models (LLMs), software, etc., with image recognition capabilities. By testing the accuracy of different LLMs/AI Agents in recognizing the calorie content of ingredients in images, comparing the deviation between the actual calorie value and the tested value of the ingredients in the image, and analyzing the image recognition capabilities of the corresponding devices or software, this dataset can be fed into the enterprise's own LLMs for training the image recognition capabilities of LLMs or AI tools.
1. Calculate the actual calorie value Q1 of food in different experiments according to the food calorie table published by the Chinese Nutrition Society: Q1 = m1/100 * q (unit: kcal); where q is the calorie per 100 grams of the corresponding food (unit: kcal), which can be obtained by looking up the table; m1 is the actual weight of the experimental food.
2. Take photos of the food in the experiment, then send the images containing the food to different LLMs/AI Agents. The LLMs/AI Agents will identify and determine the food category, the number of identified food items, and the identified food weight m2 in the image, and simultaneously provide the tested calorie value Q2 of the food in the image. Calculate the per-100-gram calorie q1 (unit: kcal) of the food during the LLM/AI Agent's calorie recognition process: q1 = 100 * Q2 / m2. Record whether the LLM/AI Agent's recognition of the food category in the image is accurate (the output name should be basically consistent, e.g., chiffon cake and sponge cake can be recognized as cake).
3. Calculate the food weight recognition accuracy A of the LLM/AI Agent for the image: A = (m2 - m1)/m1; then judge whether the weight recognition is accurate. If the absolute value of A (|A|) ≤ 50%, the "weight recognition accuracy result" corresponding to this set of data is "Yes"; if |A| > 50%, the corresponding "weight recognition accuracy result" is "No".

搜集汇总

背景与挑战

背景概述

该数据集包含1280条记录，用于评估多个APP或大模型通过图像识别食物重量和热量的准确度，涵盖食物种类、实际与识别数据对比及准确率计算。数据集适用于软硬件制造企业，通过分析偏差优化图像识别模型训练，提升热量识别能力。数据以Excel格式存储，基于标准算法规则进行热量值和重量识别准确率的评估。