Progress for week 42 (2018)
From Robin
(Difference between revisions)
(→Jonas S. Waaler) |
|||
Line 4: | Line 4: | ||
* Develop convolutional neural networks and train | * Develop convolutional neural networks and train | ||
* Compare dense and convolutional, and write about it | * Compare dense and convolutional, and write about it | ||
+ | |||
+ | === Accounting === | ||
+ | * Dense neural networks and convolutional neural networks are trained with various parameters | ||
+ | ** Results are OK | ||
+ | ** Need huge amount of trainable parameters to get OK results using Dense neural networks (ca. 33 times more compared to convolutional) | ||
== David Kolden == | == David Kolden == |
Revision as of 08:15, 19 October 2018
Contents |
Jonas S. Waaler
Budget
- Continue training dense neural networks of different sizes
- Develop convolutional neural networks and train
- Compare dense and convolutional, and write about it
Accounting
- Dense neural networks and convolutional neural networks are trained with various parameters
- Results are OK
- Need huge amount of trainable parameters to get OK results using Dense neural networks (ca. 33 times more compared to convolutional)
David Kolden
Budget
- Implement D2CO algorithm in a ROS node.
- Create Gazebo simulation model of a camera.
- Find a way to create a Gazebo model of a part that should be registered by the algorithm.
- Find a way to calculate inertia from a CAD file.
- Finish introduction of the essay.
- Finish the robotic assembly section of essay.
Accounting
- Personal page created.
- Camera model implemented in ROS/Gazebo
- Imported a part object into ROS/Gazebo from CAD file.
- Found out you can calculate inertia tensors with MeshLab
- Created CameraObjectLocalizer class, but have not compiled with D2CO yet.
Anders Rønningstad
Budget
- Read about and understand how to use mathematical optimization packages in python.
- Implement mathematical optimization in program.
- Hopefully start to run tests.
Accounting
- Found out how to implement the optimization.
- Ran some tests, with bad results.
- Tried to implement the optimization using bruteforce, but computation was to slow for it to be useful.
- Started to implement a self made method, which combines bruteforce and some assumptions.
- Ran som tests using a simple version of the new method, with some better results.