| Proj No. | A3064-251 |
| Title | Fine-grained Image Recognition by Deep Learnin |
| Summary | Fine-grained image recognition is to classify the images with subtle distinctions like birds, dogs, plants, aircrafts, and cars, etc. Samples belonging to different subclasses (e.g. Huskies and Alaska dogs) but the same class(e.g. dog) are usually with similar appearances. In addition, large appearance variances in scale, pose, lighting, and viewpoint of objects within the same subordinate category further add to the complexity of the problem. This project aims to study the basic concepts of CNN(Convolutional Neural Network) based fine-grained image recognition, and utilize the novel algorithms to learn discriminative features for distinguishing the similar fine-grained subclasses. Firstly, visual attention algorithm is applied to localize the discriminative local parts of the input images as localizing and extracting discriminative features not only from global images but also from distinctive parts play a crucial role in improving fine-grained image recognition performance. Secondly, an end-to-end feature learning network will be constructed to learn features from the localized parts. Then features of these discriminative parts will be fused to better recognize the fine-grained images. The experiment results of mostly used fine-grained datasets(e.g. CUB200, Stanford Dogs) will be analysed, and the effect of the attention and feature learning module will also be evaluated. |
| Supervisor | A/P Jiang Xudong (Loc:S1 > S1 B1C > S1 B1C 105, Ext: +65 67905018) |
| Co-Supervisor | - |
| RI Co-Supervisor | - |
| Lab | Centre for Advanced Robotics Technology Innovation (CARTIN) (Loc: S2.1-B3-01) |
| Single/Group: | Single |
| Area: | Intelligent Systems and Control Engineering |
| ISP/RI/SMP/SCP?: |