Re: rounding issue with RTCRtpEncodingParameters scales?

I think that's the same thing I'm saying, but with a different way of
thinking about it (and in reverse order from small to large).  So I think
we're in agreement.


On Fri, Jul 18, 2014 at 3:22 PM, Bernard Aboba <Bernard.Aboba@microsoft.com>
wrote:

>  Right now the resolutionScale is defined as cumulative, so if you had
> three layers and you wanted to encode full resolution with all three, it
> would be:
>
>  .25, .5 and 1.0
>
>  S0 = one quarter resolution
> S1 = one half resolution (add a pixel of width and height for each S0
> pixel)
> S2 = full resolution (add a pixel of width and height for each S1 pixel)
>
> On Jul 18, 2014, at 2:45 PM, "Peter Thatcher" <pthatcher@google.com>
> wrote:
>
>   I think each layer is a ratio of the input.  So, if you want 30%, 20%,
> 10%, you'd use 0.3, 0.2, 0.1 (not 0.3, .66666, .33333)
>
>
> On Fri, Jul 18, 2014 at 2:37 PM, Robin Raymond <robin@hookflash.com>
> wrote:
>
>>
>>  I still don’t think I fully understand exactly what those values need
>> to be set to in order to setup the correct ratios. For the sake of clarity,
>> if the base layer is scaled resolution of 30%, and you had 3 geometric
>> spatial layers, what exactly would be the value of the encoding parameters
>> for each layer?
>>
>>  --
>> Robin Raymond
>>
>>    On July 18, 2014 at 3:09:57 PM, Peter Thatcher (pthatcher@google.com)
>> wrote:
>>
>>   I think we can until we run into real implementation issues.  I have a
>> feeling it won't be a big deal.
>>
>>
>> On Thu, Jul 17, 2014 at 12:37 PM, Robin Raymond <robin@hookflash.com>
>> wrote:
>>
>>>
>>>  dictionary RTCRtpEncodingParameters {
>>>     //...
>>>      double              resolutionScale;
>>>     double              framerateScale;
>>>  }
>>>
>>>  These are floats with relative values. The trouble is that there are
>>> fixed geometric relations. It must be 1.0, 0.5, 0.25, 0.125, etc for
>>> resolution scaling…
>>>
>>>  Maybe this isn’t an issue but I’m a bit concerned that float rounding
>>> issues might cause us problems.
>>>
>>>  For example, if you start at a base scale of 0.3 (i.e. resolution is
>>> 0.3 of source), are the layerings relative to 0.3, i.e. (1/1 * 0.3), (1/2 *
>>> 0.3), (1/4 * 0.3) ?
>>>
>>>  or would the scale be like this:
>>> 0.3, 1/1, 1/2, 1/4, 1/8, etc?
>>>
>>>  Also if people use dividers vs fixed value like: 1/8 vs 0.125, could
>>> we end up in situations where the values almost match but don’t quite due
>>> to float rounding problems?
>>>
>>>
>>>  Just a concern…
>>>
>>>  Robin
>>>
>>>
>>
>