Back to Search Document details
39th Meeting: Daejeon, KR, June 2025 2025-06-30 04:27
EE1-2.1: Improved VLOP Attention with SIMD acceleration
Authors: Yun Li Qualcomm Muhammed Coban Qualcomm Marta Karczewicz Qualcomm L. Kerofsky Qualcomm
Abstract
VLOP1 was derived from LOP in JVET-AH0051. VLOP3 with earlier cropping and reduced inputs were integrated in JVET-AJ0054 and JVET-AJ0066. In JVET-AL0166, an improved residual groups and spatial attention with VLOP3 was proposed with a significant coding gain.
Meeting Notes
JVET-AM0135 EE1-2.1: Improved VLOP Attention with SIMD acceleration [Y. Li, M. Coban, M. Karczewicz, L. Kerofsky (Qualcomm)]
Cited:
JVET-AM0272 X JVET-AQ0172
ARCHIVE :
JVET-AM0135-v2.docx JVET-AM0135-v2_clean.docx JVET-AM0135_VLOP_complexity.xlsx JVET-AM0135_vs_VLOP3_NNVC13_v1.xlsm JVET-AM0135_vs_VLOP3_NNVC13_with_improved_SIMD_for_both_anchor_and_test_v1.xlsm JVET-AM0135_vs_VLOP3_NNVC13_with_improved_SIMD_v1.xlsm JVET-AM0135_vs_VTM_NNVC13_v1.xlsm
Decisions
JVET-AM0135 EE1-2.1: Improved VLOP Attention with SIMD acceleration [Y. Li, M. Coban, M. Karczewicz, L. Kerofsky (Qualcomm)]
Citation