JVET-AJ0087 Non-EE2: Block vector guided DIMD [L. Zhang, Y. Yu, H. Yu, D. Wang (OPPO)]

This contribution proposes a block-vector guided DIMD method (BVG-DIMD) that uses block vectors to locate reference areas, construct a histogram of gradients for the samples in the reference areas, and derive intra modes based on the histogram. In the BVG-DIMD mode, the prediction of the block is obtained by using these intra modes. The proposed method is implemented on top of ECM-14.0, and the simulation results are reported as follows:

AI: { -0.14% Y, 0.00% U, 0.05% V, 104.4% EncT, 106.0% DecT }

RA: { xx% Y, xx% U, xx% V, xx% EncT, xx% DecT }

It was commented that the reported tradeoff between compression and encoder/decoder runtimes is far from being attractive. Also, some of the gain may be decreased due to the new adoptions in the DIMD area. Proponent is informed that the chances for adoption would be low, unless the performance is significantly increased.

Interest was expressed by other experts – it was agreed to investigate this in an EE.