「利用者:Tianwei37/Report & Blog/Week 3」の版間の差分
(→What I did) |
細 (1版 をインポートしました) |
(相違点なし)
|
2018年6月29日 (金) 06:17時点における最新版
What I did
- refine correspondence data structure according to Sergey’s suggestion
- revise BKE_tracking_correspondence_add accordingly
- add BKE_tracking_multiview_reconstruction_check which checks conditions for multi-view reconstruction
- Clear up my mind about BLI_list, BLI_array, BLI_buffer. Stuck by a bug for a long time, then It turned out that the bug is caused by passing a data structure (MovieClip) without *next and *prev pointers into a blender_list.
- Refine data structures for multi-view reconstruction
- Extend my report into a documentation about multi view reconstruction at here: https://wiki.blender.org/index.php/User:Tianwei37/Report_%26_Blog/Multi-view_Reconstruction_Documentation
- manage BKE_tracking_multiview_reconstruction_context_free as several data structures have been changed to arrays.
- convert the simple_pipeline code to autotrack code. Basically the entire reconstruction pipeline should now be based on the new autotrack pipeline.
What I will do next week
Working towards the tentative goal of week4-6 in my GSoC proposal, Connect the front side UI with the back-side SfM engine. More specifically, I will continue to focus on the libmv autotrack pipeline next week. Currently libmv/autotrack is immature in that several header files are designed without the actual *.cc implementation. I will basically learn from libmv/simple_pipeline and migrate everything needed for multi-view reconstruction to libmv/autotrack.
Questions
1. What is fundamental difference between autotrack and simple_pipeline? From my point of view, autotrack is based on mv::marker which knows what clip it comes from, while simple_pipeline is based on libmv::marker which is agnostic about the clip. Also autotrack has data structures that have more details than their counterparts in simple_pipeline, like libmv_Marker. It knows the marker source, like whether it is specified manually, or detected automatically, which can lead to fine-grained controls in the algorithm implementations. Is there anything else that I have missed and should attend to?
2. Does anyone know how to delete a wiki page? I accidentally create one more and want it deleted.