We've been developing a product containing ADV212 codec and ran into the following issue - the default quantization steps are too crude for our goals at 12 BPP, however I have not found any place in the documentation which describes what quantization steps can be used. Essentially the question is what are the minimal quantization steps which will not cause the ADV212 to crash.
We were trying to use the quantization steps QStep=0.0005 (in the notation used by Kakadu library) however the ADV212 crashes when the image is too noisy - close to completely random data. The tile size itself is well below limits - we've tried 1650x137 and 1650x68, the BPP is 12, 5 DWT levels (6 resolutions), 9x7 irreversible kernel.
I've seen that excessive tile sizes can lead to ADV firmware crashes, due to memory exhaustion, such as http://ez.analog.com/message/24508#24508, however our configuration appears to be much below the limits and the only reason I can see for problems is if for some reason the amount on data produced by our quantization scheme before rate control exceeds some internal ADV limit.
So information regarding any of the following will help:
- How does one detect an ADV212 crash? Is it possible to determine the cause of its crash?
- What are the officially supported limitations of Quantization steps?