October 4th, 2012
I’ve cut the stable branch off unstable yesterday, and a r0.2-rc branch as well. This marks the 0.2 version of ccv finally here.
ccv 0.2 expects to be a bug-fix release, and should address several “obvious” problems have after the first release. Thus, two important issues get addressed in this release:
1). thread-safety. The first release version of ccv is not thread-safe when enabled the application-wide cache, because access and mutation on this cache is not thread-safe. In this release, the application-wide cache becomes thread local, thus, have its thread-safety now.
2). read in-memory data. It is always possible to create a ccv_dense_matrix_t yourself and fill it out with in-memory data, but the lacking of canonical way to read in-memory data into ccv seems problematic in the community. In this release, ccv_read will take more options, and one of them is to read in-memory self-describe data (such as in-memory JPEG or PNG), another is to read RGB plain formatted bytes directly. read more.
In 0.2, DPM finally can train multi-components model (a.k.a. the “real” mixture model), and I’ve included a car model for you to try out:
Largely still work-in-progress, the SWT included a swtcreate tool for you to tune SWT parameters to fit your use-cases, and the text segregation gets better too:
The last 3 months have been “bumpy” for me, and after the summer school in Berkeley, I should be able to release ccv more regularly. Hope you enjoy.comments powered by Disqus