[Soc-2013-dev] Weekly Report #8 VSE

Alexandr Kuznetsov kuzsasha at gmail.com
Mon Aug 12 06:54:11 CEST 2013


Hi Peter:

> GSoC is supposed to be an open development process, so please commit
> your work at the*very*  least every three days.

Sorry about that. I changed things around, so it was a little messy. 
Expect commits from me every day starting tomorrow.

> If you found a bug, fix it.
>
> Skipping pixels makes sense for preview quality, for output rendering
> it's obviously a bad idea(tm).

It seems that the problem is for the whole blender. In fact even 
Photoshop has the same problem with bilinear resampling (but save for 
web works correctly). Bicubic should be also reproducible with more 
radical size reduction. The fact is, we just calculate the point of the 
new pixel and then interpolate from available data. This method doesn't 
take into the account that new pixel essentially represents 16 old 
pixels while it is interpolated from just 4 (bilinear case).
To solve this I can extract the scale and apply the transformation and 
then resize with more precise downsampling algorithm.

Best,
Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/soc-2013-dev/attachments/20130812/280cc739/attachment.htm 


More information about the Soc-2013-dev mailing list