Search Results for "JVET-AA0113"
Found 4 document(s)
Search documents
Use number, keyword, author, or MPEG number. Filter by meeting when needed.
Not signed in yet.
Abstract
This contribution presents an encoder optimization technique on top of the deep learning-based in-loop filtering method combined with deblocking in EE1-1.6. The proposed method is asserted to improve the compression efficiency at encoder by introducing CNN-based filtering and deblocking into the rate-distortion optimization (RDO) process. In the RDO process, CNN-based filtering is implemented with SADL using fixed point-based calculation. Compared with VTM11.0_nnvc, the proposed method reportedly shows on average {x%, x%, x%}, { x%, x%, x%}, and {x%, x%, x%} BD-rate reductions for {Y, Cb, Cr} components, under AI, RA, and LDB configurations, respectively.
References:
Citation
Abstract
This contribution reports the crosscheck results for JVET-AA0113 [], which introduces CNN-based filtering and deblocking into the rate-distortion process. On the top of EE1.6, the CNN-based filtering is implemented on the SADL library using fixed-point calculation. It is reported that the partial simulation results are exactly match to the results provided by the proponents under random access configuration.
References:
Citation
Abstract
This contribution reports on a training and inference crosscheck of EE1-1.6, proposed at the last meeting as JVET-AA0113 and presented at this meeting in JVET-AB0068. Two networks, luma-intra-RDO and luma-inter-RDO, were trained from scratch using training data extracted using the supplied software. The networks were converted to integer SADL and then evaluated against NNVC-2.0 and NCS 1.0 filter set 1. In the evaluation, the models used for actual NN loop filtering were the same as the one in JVET-AA0113, only the luma-intra-RDO and luma-inter-RDO were changed to the retrained versions. The results using the retrained version compared to JVET-AB0068 are reported to be:
References:
ARCHIVE
:
._.DS_Store
._JVET-AB0137-v1.docx
._JVET-AB0137-v2.docx
._JVET-AB0137-v2_clean.docx
._JVET-AB0137_crosscheck_of_AA0113_EE1-1.6_vs_NCS-1.0_filter1.xlsm
._JVET-AB0137_crosscheck_of_AA0113_EE1-1.6_vs_NNVC-2.0.xlsm
._JVET_AB0137_crosscheck_of_AA0113_EE1-1.6_vs_JVET-AA0113.xlsm
._JVET_AB0137_crosscheck_of_AA0113_EE1-1.6_vs_JVET-AB0068.xlsm
.DS_Store
JVET-AB0137-v1.docx
JVET-AB0137-v2.docx
JVET-AB0137-v2_clean.docx
JVET-AB0137_crosscheck_of_AA0113_EE1-1.6_vs_NCS-1.0_filter1.xlsm
JVET-AB0137_crosscheck_of_AA0113_EE1-1.6_vs_NNVC-2.0.xlsm
JVET_AB0137_crosscheck_of_AA0113_EE1-1.6_vs_JVET-AA0113.xlsm
JVET_AB0137_crosscheck_of_AA0113_EE1-1.6_vs_JVET-AB0068.xlsm
Citation
Abstract not available in document
Loading full meeting notes...
JVET-AA0247 BoG on Neural Network Video Coding (NNVC) [A. Segall, E. Alshina]
This is a report of activities from the BoG on Neural Network Video Coding (NNVC). The BoG held the following meetings during the 27th JVET meeting:
July 19 – 05:00-7:00 UTC
July 19 – 15:35-17:35 UTC
July 20 – 15:20-17:00 UTC
The BoG activity is summarized as:
The BoG reviewed the training crosschecks, including:
Summarizing key takeaways as:
The crosschecks took on the order of weeks for training.
The crosschecks took on the order of weeks for data generation. In some cases, this was further increased due to data transfer.
The training scripts provided by proponents did not work in all compute environments
The crosschecking effort would be improved by defining the naming convention for common training data, such as the number of coded frames and QPs.
Re-training of the model leads to a worst-case performance d...
Decisions
See under section 4.8.
Citation