本数据集包含了101 种食品类别的图像数据集,共有 101,000 张图像,平均每个类别拥有 250 张测试图像和 750 张训练图 像。训练图像未经过数据清洗。所有图像都已经重新进行了尺寸缩放,最大边长达到了 512 像素。
In this paper we address the problem of automatically recognizing pictured dishes. To this end, we introduce a novel method tomine discriminative parts using Random Forests (rf), which allows usto mine for parts simultaneously for all classes and to share knowledgeamong them. To improve efficiency of mining and classification, we onlyconsider patches that are aligned with image superpixels, which we callcomponents. To measure the performance of our rf component miningfor food recognition, we introduce a novel and challenging dataset of101 food categories, with 101’000 images. With an average accuracy of50.76%, our model outperforms alternative classification methods exceptfor cnn, including svm classification on Improved Fisher Vectors andexisting discriminative part-mining algorithms by 11.88% and 8.13%, respectively. On the challenging mit-Indoor dataset, our method comparesnicely to other s-o-a component-based classification methods.
译:
本文讨论了图像菜品的自动识别问题。为此,我们引入了一种新的方法使用随机森林(rf)挖掘区分部分,这允许我们为所有课程同时挖掘零件并共享知识其中之一。为了提高挖掘和分类的效率,我们只考虑与图像超级像素对齐的面片,我们称之为组件。来衡量我们的射频分量挖掘的性能对于食品识别,我们引入了一个新颖且具有挑战性的数据集101种食物,有101000张图片。平均精度为50.76%,我们的模型优于其他分类方法,除了对于cnn,包括基于改进Fisher向量的svm分类和现有的判别部分挖掘算法分别提高了11.88%和8.13%。在具有挑战性的麻省理工学院室内数据集上,我们的方法进行了比较与其他基于s-o-a组件的分类方法相比,效果很好。
大家可以到官网地址下载数据集,我自己也在百度网盘分享了一份。可关注本人公众号,回复“2020082001”获取下载链接。
只要自己有时间,都尽量写写文章,与大家交流分享。
本人公众号:
CSDN博客地址:https://blog.csdn.net/ispeasant