Back to Search Document details
29th Meeting: by teleconference, DE, January 2023 2023-01-18 13:58
Crosscheck of EE1-3.2 (JVET-AC0116 : neural network-based intra prediction with learned mapping to VVC intra prediction modes)
Authors: Mitra Damghanian Ericsson J. Ström Ericsson
Abstract
This contribution reports on the training and inference crosscheck of EE1-3.2 JVET-AC0116, proposed at the last meeting as JVET-AB0149. The neural network-based intra prediction mode comprises 7 different fully connected neural networks. The training procedure of the proponents consists of four cycles of data generation and training. In this crosscheck, this has been broken into three steps, each with an evaluation point at the end. At each evaluation point, the performance was measured using both the proponent’s models and the models trained from the crosscheck. The first step started from a random network initialization and the two other steps were trained starting from existing networks provided by the proponent. All the training steps were done using training data extracted using the supplied software. The three evaluation points resulted in the following BDR figures:
Meeting Notes
JVET-AC0270 Crosscheck of EE1-3.2 (JVET-AC0116: neural network-based intra prediction with learned mapping to VVC intra prediction modes) [M. Damghanian, J. Ström (Ericsson)] [late]

This document is about training crosscheck. It was confirmed during session 27 that the crosscheck was completed successfully (version 4 of JVET-AC0270).

References:
JVET-AC0116
ARCHIVE :
._.DS_Store ._EE1-3.2.1_crosscheckers_models_plus_NNLF0_over_NNVC3_without_NNLF.xlsm ._EE1-3.2.1_proponents_models_plus_NNLF0_over_NNVC3_without_NNLF.xlsm ._EE1-3.2.2_crosscheckers_models_plus_NNLF1_over_NNVC3_without_NNLF.xlsm ._EE1-3.2.2_proponents_models_plus_NNLF1_over_NNVC3_without_NNLF.xlsm ._Evaluation_point_1_crosscheckers_models.xlsm ._Evaluation_point_1_proponents_models.xlsm ._Evaluation_point_2_crosscheckers_models.xlsm ._Evaluation_point_2_proponents_models.xlsm ._Evaluation_point_3_crosscheckers_models.xlsm ._Evaluation_point_3_proponents_models.xlsm ._JVET-AC0270_v1.docx ._JVET-AC0270_v3.docx ._JVET-AC0270_v4.docx ._JVET-AC0270_v4_clean.docx ._~$ET-AC0270_v3_clean.docx ._~$Evaluation_point_1_proponents_models.xlsm ._~$Evaluation_point_3_crosscheckers_models.xlsm ._~$Evaluation_point_3_proponents_models.xlsm .DS_Store EE1-3.2.1_crosscheckers_models_plus_NNLF0_over_NNVC3_without_NNLF.xlsm EE1-3.2.1_proponents_models_plus_NNLF0_over_NNVC3_without_NNLF.xlsm EE1-3.2.2_crosscheckers_models_plus_NNLF1_over_NNVC3_without_NNLF.xlsm EE1-3.2.2_proponents_models_plus_NNLF1_over_NNVC3_without_NNLF.xlsm Evaluation_point_1_crosscheckers_models.xlsm Evaluation_point_1_proponents_models.xlsm Evaluation_point_2_crosscheckers_models.xlsm Evaluation_point_2_proponents_models.xlsm Evaluation_point_3_crosscheckers_models.xlsm Evaluation_point_3_proponents_models.xlsm JVET-AC0270_v1.docx JVET-AC0270_v3.docx JVET-AC0270_v4.docx JVET-AC0270_v4_clean.docx ~$ET-AC0270_v3_clean.docx ~$Evaluation_point_1_proponents_models.xlsm ~$Evaluation_point_3_crosscheckers_models.xlsm ~$Evaluation_point_3_proponents_models.xlsm
Decisions
This document is about training crosscheck. It was confirmed during session 27 that the crosscheck was completed successfully (version 4 of JVET-AC0270).
Citation