A Deep Learning Approach to Analyze the Relationship Between Gender, Height, Weight, and Basal Metabolic Rate from Face Images

In response to existing limitations, a novel convolutional neural network (CNN) model tailored for facial attribute estimation is introduced in this study. The methodology encompasses meticulous data preprocessing techniques, including dataset splitting, resizing, cleaning, augmentation, and cropping, to ensure the integrity of the dataset. Through rigorous real-world testing, remarkable accuracy in predicting gender, height, weight, and basal metabolic rate (BMR) from facial photos is demonstrated by our CNN architecture. Notably, unlike conventional approaches where BMR is directly estimated from facial images, three output neurons are employed by our model to simultaneously predict gender, height, and weight, which are sub-sequently utilized for BMR calculation. A primary contribution is made through the development of a comprehensive methodology that advances the reliability of facial attribute estimation by achieving an outstanding accuracy of 98.50% on our dataset named Face-ete and a commendable accuracy of 88.29% on another dataset. Our model outperforms the second-best model by percentage relative improvement factors of 2.19 and 4.28 in terms of BMR prediction accuracy for the “Faceere” and the other dataset, respectively. This work represents a significant step forward in the field of facial attribute estimation.