[CinCV TNG] RFC:Next version

Einar R√ľnkaru einarrunkaru at gmail.com
Thu May 21 19:19:36 CEST 2015


Hi

On 05/20/2015 09:29 PM, Johannes Sixt wrote:
> Am 20.05.2015 um 19:51 schrieb Nicola Ferralis:
>> Not sure if this is where the problem is... But one notable difference
>> between the MaskUnit::do_feather in CV 2.2 (where it seems to be broken)
>> and HV 4.6 (where it seems to work) is the casting of feather. In 4.6:
>>
>> void MaskUnit::do_feather(VFrame *output,
>>                            VFrame *input,
>>                            double feather,
>>                            int start_y,
>>                            int end_y,
>>                            int start_x,
>>                            int end_x)
>>
>> in CV:
>>
>> void MaskUnit::do_feather(VFrame *output,
>>      VFrame *input,
>>      float feather,
>>      int start_out,
>>      int end_out)
>>
>> This is percolated through the code with other variables been casted as
>> double rather than floats in MaskUnit::do_feather.
>
> It's not impossible. It is barely an overflow as the numbers do not grow
> ad infinitum. On the other hand, there are a number additions and
> subtractions, and it is possible that numeric imprecision causes the
> observed problems.
>

I see one more difference: there are 4 parameters after feather in HV vs 
2 parameters in CV. I suspect that there are more changes in HV. This 
means that the real fix is not small and simple.

Small and simple fix is to restrict feather to 0..40.

My idea is to release next version before we start to merge from HV.

Einar


More information about the Cinelerra mailing list