Practitioner Talk Data Science: Image Aesthetic Assessment
One of the biggest challenge facing any content driven platform is to showcase high quality images to users. Our work focuses on the automatic assessment of image aesthetics using Convolutional Neural Network (ConvNet). The constraint of fixed size input in ConvNet compromises the aesthetics of original image. We incorporated the Adaptive Spatial Pooling technique to address this problem by fixing the layer size just before the Fully Connected Layers in ConvNet. This preserves the original image composition and enables the model to learn aesthetic features without any transformations. NLP classification and sequence pattern recognition problems also falls under the purview of this work.