T71 kirjoitti:niffe kirjoitti:Ohjelmallisilla dynamiikan manipuloinneilla on ilmeisesti tradeoffinsa väripuolella, sillä DxOmarkin mukaan A7R3:n color sensitivity on rommannut melkein aukon verran verrattuna A7R2:een.
Missähän näistä R2, R3 ja III väreistä voisi lukea enemmän? Kiinnostaa saisiko sellaisen kameran, jossa olisi luonnollisemmat värit, eli kehittyykö Sony hiljalleen siihen suuntaan vai miten tämä color science näissä menee.
Hyshys, ei täällä tai missään muillakaan kamerafoorumeilla saa keskustella muista kennon ominaisuuksista kuin korkeiden herkkyyksien dynamiikasta ja kohinoista; väritoisto ja pikselitason terävyys ovat kiellettyjä aiheita!
Mutta jos aihe kiinnostaa, niin linkin takana nimimerkki
theSuede kertoo asiaa (Kyllä, samanlainen RGB-filtteri siellä Sonynkin kennolla on):
http://www.fredmiranda.com/forum/topic/1013683/2
theSuede - FM.com kirjoitti:The 5D2 is both more AND less like the human eye when you look at it from a more controlled, dispassionate PoV.
The difference between them is that in the generations starting with 50D Canon has changed the classic "R,G,B" type of Bayer colour filters into "orange, warm green, blue". This is, if you do it well, and control your filter parameters well a very good thing for most intents and purposes.
A few things makes the solution less-than-optimal though. When you design a colour filter set, you only have a fixed amount of materials to work with. The materials you use MUST be able to function in the process that physically makes the filters (we're talking about depositing +/-2% amount material onto a very well defined square about 5x5µm large here...) and it has to be non-fade, non-sensitive to certain material used in the sensor surface and so on.
So, to get as close to target as they can, Canon has used a mix of material that sometimes puts the camera into trouble. Especially the "double-hump" in the orange channel in contemporary models can wreak havoc with colour accuracy in the range red>green.
But one does have to separate "accurate colour" from "pleasing colour". Even though the 5D2 and the likes of it have some very real problems (very red orange, and purples are very hard to get right - even in perfect light) it also has some positive effects.
When the light goes crazy (very low K temperatures, fluorescent lights and so on) the 5D2 keeps on top of trying to get skin-tone "about the right mid-orange", and "about mid-high saturation". This is not in any way "accurate", but it's easy to work with.
In good, balanced light spectras (daylight, studio flash) the 5D classic is miles ahead in colour definition and hue accuracy though. You get a much higher "hue-resolution" than with a 5D2. In landscaping, this might be seen as a lot higher colour definition - two trees standing next to each other have a different green colour base. Parts of foliage are more saturated with either phycocyanin, carotene, or xanthophyll chlorofyll base mixes, making them all slightly DIFFERENT green compared to the foliage parts right next to it. Some cameras can differentiate between those (very similar) colours, some just can't.
So what the 5D2 gives you is "average skintone", almost no matter what the circumstances are. This might be seen as good or bad depending on what you prefer. It also introduces some other very serious problems - like the magenta/green chroma noise caused by the unusually large amount of saturation amplification needed to make the raw file red>green range look like real colours. This is a very real "Canon problem", and the colour filter choices they've made is what causes it.
theSuede - FM.com kirjoitti:
If you want to compare filter sharpness on the DxO site, the easiest way is to check the NW>SE direction diagonal line of the raw>sRGB matrix.
5D2: 2.25 - 1.62 - 1.43
5Dc: 1.96 - 1.53 - 1.31
1Ds3: 2.02 - 1.50 - 1.31
Higher numbers mean that the colour saturation has to be amplified more. The number do in no way guarantee that the colours are accurate, it just states what you have to do to get the primary channels' saturation right. Higher numbers does hence mean very close to [number^2] more chroma noise.
But the hue accuracy and hue resolution isn't just about filter sharpness, it's also about how far apart you space the peaks. Peaks further apart: Better hue resolution, worse resistance to metamerism failure.
DxO-mittauksiin (tässä esimerkkinä 1Ds III) linkki alla:
https://www.dxomark.com/Cameras/Canon/E ... asurements
Edit. Linkki ei vie suoraan Color Response-välilehden alle.
Edit2. Pari FM-ketjua lisää, missä aihetta käsitellään:
http://www.fredmiranda.com/forum/topic/1234124/2
http://www.fredmiranda.com/forum/topic/1374664/5