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:
- Brief Report on existing algorithms for object detection, which can be used as guidelines for future contributions in OpenDetection. Something on these lines
- Basic Utility functions for Caffe interfacing.
- Implementation of base trainer class for CNN based methods.
- Implementation of base trainer class for CNN based classification methods.
- Scripts for data preparation for classification.
- Implementation of base detector class for CNN based detection methods.
- Implementation of classifiers methods for various CNN architectures (Alexenet, VGG-16, ResNet).
- Implementation of detector method based on pre-trained detector methods (faster-RCNN, YOLO).