JVET-F0044 AHG5: Performance evaluation of Adaptive Search Range on JEM-5.0.1 [Y. Yasugi, T. Ikai (Sharp)]

This contribution reports the performance of Adaptive Search Range (ASR) on JEM-5.0.1. The authors tested the performance on JEM-5.0.1 using several combinations of minimum search range (MinSearchWindow) and maximum search range (SearchRange), where the minimum search range setting implementation was ported from HM 16-15. It is reported that the following configurations (MinSearchWindow_SearchRange) can be found as good operation points in terms of performance and encoding time balance: 128_256, 96_384 and 64_512.

From the results shown here, the combination 96_384 gives best tradeoff. However, the CTC of JEM and HM should be kept aligned, and HM results have not been investigated yet. HM results were provided in an update of the document, and indicated a minor improvement with a minor reduction of runtime, such that it seems to be beneficial for HM as well. This was reported to JCT-VC in JCTVC-AA0043.

As it would otherwise be necessary to generate new HM anchors for the CfE, it would be undesirable to implement this change at the current meeting. The change to the combination 96_384 adaptive search range should be implemented in CTC by the July meeting in both HM and JEM CTC.