Abstract:
This paper introduces versatile filters to build efficient convolutional neural networks used in many visual recognition tasks. Several methods have been developed to learn compact neural networks for efficient deep learning on cheap hardware. These studies investigate small, sparse, or quantized filters to reduce filter size. We view filters additively. Binary masks can generate multiple secondary filters from a primary filter. These secondary filters inherit from the primary filter without taking up more storage, but when unfolded in computation, they can greatly improve the filter by integrating information from different receptive fields. We also study channel-perspective versatile filters. Binary masks can be modified for orthogonal primary filters. We analyze network complexity theoretically and introduce an efficient convolution scheme. Our versatile filters perform similarly to original filters on benchmark datasets and neural networks, but with less memory and computation.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
