Tianhan:
This week I did a lot of benchmark on my trained hand detection model. I also trained another model using faster rcnn as backbone to compare the outcome. It turns out that the model trained with faster rcnn was actually too slow for our purpose (less than 1 fps), and the origin one trained with SSD is much better (around 5-6 fps).
I also continued to write the server side code in order to deploy our program as a webapp.
Xinyu Zhao:
This week we kept integrating and I attempted to solve the problem of hand interfering with AR effect display and it turned out in order for this to work we might have to use a completely different method with ML and we’ll have to search for table edge image database to train. So we probably would just stick to the original method. Also started to work on the front end of the webapp and learned about Django.
Team:
This week we managed to combine every part together. The next target would be to finish the webapp frontend and backend, and polish more on the AR part.
0 Comments