Search for Finer Attention Details for Fine-Grained Image Classification
Abstract
The task of general object classification via images has been studied extensively. It involves
distinguishing very different object categories like a dog or a cat. On the other hand, the task of
fine-grained classification deals with the recognition of images having subtle visual differences
among the classes or categories. The marginal visual difference between different classes in fine-grained images makes this very task harder.
The work of this thesis is inspired by how the human visual system looks for fine attention details to recognize an object in the image. Our brain is trained to look for some particular fine discriminative details by repetitively scanning through the image. Through our work, we tried to focus on these marginal differences to extract more representative latent features via deep learning models. Similar to human vision, our network recurrently focuses on the parts of images to spot small discriminative parts among the classes. Moreover, we show through interpretability techniques how our network focus changes from coarser to finer details. Our network uses only image-level labels and does not need bounding box/part annotation information to spot these changes. Further, the simplicity of our network makes it an easy plug-n-play module increasing its usability in other applications.
Citation
Shroff, Prateek (2020). Search for Finer Attention Details for Fine-Grained Image Classification. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /192579.