Open Detection

Mentors: Student: Gautam Malu

<p>Adding CNN based classifiers and dectors to the OpenDetection framework. This requires implementation of training methods and detection methods for both classifiers and detectors with Caffe backend. The main deliverables are:</p> <ul> <li>Brief Report on existing algorithms for object detection, which can be used as guidelines for future contributions in OpenDetection. <a href="https://github.com/gautamMalu/opendetection/blob/algo_report/ObjectDetection.md" target="_blank">Something on these lines </a></li> <li>Basic Utility functions for Caffe interfacing.</li> <li>Implementation of base trainer class for CNN based methods.</li> <li>Implementation of base trainer class for CNN based classification methods.</li> <li>Scripts for data preparation for classification.</li> <li>Implementation of base detector class for CNN based detection methods.</li> <li>Implementation of classifiers methods for various CNN architectures (Alexenet, VGG-16, ResNet).</li> <li>Implementation of detector method based on pre-trained detector methods (faster-RCNN, YOLO).</li> </ul>

Mentors: Student: Xiongtao Chen

<p>Open Detection is a standalone open source project for object detection and recognition in images and 3d point clouds. In this proposal, I first make a short report of the state-of-the-art detection/recognition techniques. To improve the framework for CNNs for 2D detection, I mainly focus on two key tasks that remain to be addressed. The first one is to implement OD interface for training and testing neural network architectures with CNN and RNN. The second is to add several recognition/detection algorithms to OpenDetection. To this end, I decompose each task into several small tasks and propose my solutions respectively, aiming to perfect the project in an elegant way. I also create milestones for each small tasks. I can work on this project full-time during this summer and I’ve equipped myself with prior knowledge and programing experiences in object detection and deep learning, which makes me competent to this project.</p>

Mentors: Student: Abhishek Kumar Annamraju

<p>The proposal mainly focuses on Convolutional Neural Network based object detection module to be integrated with the openDetection library. The library that will be used for this is caffe. The targets that will be accomplished are:1) Re-designing the library for CPU and GPU compilation modes. 2) Implement a way to invoke Caffe open source library from the OpenDetection module with a user-friendly code based way ( this will include a tinge of GUI support for instant access). 3) Implement open source guidance and codes for state-of-the art object localization problems(hypothesis generation) specifically based on selective-search and convolutional neural network (CNN) approaches. 4) Adding a ground-truth annotation tool to the module with a graphical-user-interface support. Implementing short, but effective modules like mixed-pooling, recurrent networks to the Convolutional Neural Networks Training dependent on the invoked caffe library. 5) Adding context based learning CNNs. 6) Adding user-interface to train and test CNN based classifiers and object detectors. 7)Adding documentation for the above</p>

Mentors: Student: Giacomo Dabisias

<p>The main goal of this project is to reshape the current library in such a way that it becomes easily expandable, usable and able to achieve high performance for the implemented state of the art algorithms. It is also very important to provide the library with the ability to be built cross platform among standard operating systems. After this objective, it would be good to integrate some existing state of the art algorithms. The last step consists in writing a clear documentation along with some more examples for the new library.</p>