See my Sigma mirrorless wish list.
Be careful what you wish for: the about $1199 Sigma sd Quattro-H shoots DNG as one option, but the 150MB file sizes vs 68MB for X3F are a huge (literally!) downside that certainly gives me pause about using DNG.
Shoot 10 frames and you’re at a whopping 1.5GB. Yikes. It seems to me that an X3F to DNG converter makes a lot more sense, perhaps one that could properly compress the data in the DGN.
Image dimensions as processed from DNG is 6192 x 4128 = 25.5 megapixels, which by basic math means that each pixel in the finished image consumes 5.88 bytes of storage = ~47 bits per image pixel.
The Sigma sd Quattro-H sensor is a semi true-color sensor with multiple layers. See Sigma dp Quattro Sensor Design. These Quattro sensor layers are not all full resolution as they are in the Sigma DP Merrill sensor:
- Blue layer: full spatial resolution, broad panchromatic sensitivity
- Red layer: 1/2 spatial resolution = 1/4 the data of the blue layer
- Green layer: 1/2 spatial resolution = 1/4 the data of the blue layer
Which implies 14 bits + 7 bits = 21 bits per image pixel.
Why are 47 bits per image pixel stored if the sensor data requires only 21 bits? Does the DNG contain both the X3F as well as a duplicate perhaps? If so, why won’t Sigma Photo Pro 6.5.0 process DNG files?
Sigma sd Quattro DNG file size is little different from 16-bit uncompressed TIF. But see Sigma sd Quattro-H: DNG Files Shrink by More than 50% After Adobe DNG Converter.
Bits per pixel
That’s all theory. RawDigger shows that DNG is 12-bit and that X3F is 14-bit. So bigger files with DNG and you lose 2 bits. Not so appealing and rather strange too.
It’s discouraging that using DNG will extract its pound of flesh: the 36MP Pentax K-1 in SuperRes pixel shift mode stores 4 exposures in one file, which implies up to 260MB, but it actually produces files around 191MB on average. The Pentax uses lossless compression to keep file sizes “reasonable”. Apparently the Sigma sd Quattro is not nearly so successful (3 layers and only 2/3 the pixels per layer implies something much less efficient).