Dimensions overall score 7.0

• CPUBone is a new family of vision backbone models optimized for CPU-based inference.method 95%
• Grouping convolutions and reducing kernel sizes are proposed adaptations to reduce computational cost in vision backbones for CPUs.method 95%
• Grouped MBConv (GrMBConv) has 23% less MACs than the standard MBConv block on average across five channel dimensions.result 90%
• Grouped Fused MBConv (GrFuMBConv) has 45% less MACs than the standard FuMBConv block, independent of channel dimension.result 90%
• CPUBone achieves state-of-the-art Speed-Accuracy Trade-offs (SATs) across a wide range of CPU devices.result 95%
• CPUs have limited parallelization capabilities compared to GPUs and embedded AI accelerators.technical 95%
• Existing models optimized for high parallelization capabilities often fail to translate their efficiency to CPU-based systems.limitation 95%
• CPUBone is a new family of vision backbone models optimized for CPU-based inference.method 95%
• Grouping convolutions and reducing kernel sizes are proposed adaptations to standard convolutions to reduce computational cost on CPUs.method 95%
• The GrFuMBConv block has 45% less MACs than the FuMBConv block, independent of the channel dimension.result 90%
• The GrMBConv block has 23% less MACs than the MBConv block, averaged over five channel dimensions.result 90%
• CPUBone achieves state-of-the-art Speed-Accuracy Trade-offs (SATs) across a wide range of CPU devices.result 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card