Due to JCT3V-C0137 adoption, which changes implicit motion inheritance to explit merge mode, the derivation process anyIvRefPicFlag for depth was obsolete. Specifically, the result of depth's anyIvRefPicFlag derivation is equal to the result of texture's anyIvRefPicFlag derivation.

I talked this part with the original IC proponent and they confirmed this part is obsolete and agreed that this part's removal.

While basically this change is editorial change, HTM also needs to be fixed.

At new HTM-6.2-dev integration phase, my simulation sometimes failed. I thought this was due to memory usage increase of the integrated version and checked memory usage before and after integration (HTM-6.2 and HTM-6.2-dev). Then i found the following strange behaviour (potential bug) in HTM encoder.

The memory usage of 720p was constant (2.4 G bytes in my environment) while that of 1080p varies from 2.8G bytes to 3.9Gbytes. The memory usage varies by runs!

This strange behavior occcued at leaset in HTM-6.0, HTM-6.2 and HTM-6.2-dev branch2. Please see the attached file.

Attached is my memory usage reports, which is got by qacct -j $job | grep maxvmem

I suppose this "randomly" different memory usage occured "only in 1080p" suggests potential bugs in HTM.

The decoder picture hash doesn't work in enhancement layer.

This is because encoder always send the decoded picture hash SEI with nal_layer_id equal to 0. The decoder decodes the SEI in enhancement layer but interprets that is associated with base layer.

A patch is attached.