Existing models will be implemented and results will be compared. The drawbacks will be analyzed so that the severity of these defects can be well studied. The machine learning network structure used in this project will then be proposed based on these studies.
The initially designed details will be finalized and models will be implemented. Platforms, languages, and tools used will be finalized at this stage. The design and coding style of the project will be discussed and guide the following experiments.
The main focus will be on optimisation and fine-tuning of the network structure. A renowned fact in machine learning area is that small changes of parameters can make a big difference in the final results. At this stage, previous works will be used for references to fine-tune our networks, thus achieving more desirable performance.
From March to May, the whole project will be wrapped up and results, including the finalized network structure and the end-to-end program, will be delivered. After the finalization of the network and algorithms, time will then be spent on summarization and documentation. A paper for computer vision conference can also be written as this time with description and impact of this work.