EfficientNet: For scalability

Aus HSHL Mechatronik
Version vom 11. Februar 2026, 14:04 Uhr von Ajay.paul@stud.hshl.de (Diskussion | Beiträge) (Die Seite wurde neu angelegt: „== EfficientNet: For scalability == EfficientNet use something called compound scaling. It scale the network’s width, depth, and resolution all together in a balanced way. Instead of just making the network deeper like ResNet, or wider only, EfficientNet try to find the best balance between them. Because of this, it can reach higher accuracy with less parameters and less FLOPs (Floating Point Operations).<ref name="Joshua2025">Joshua, Chidiebere & Kotsi…“)
(Unterschied) ← Nächstältere Version | Aktuelle Version (Unterschied) | Nächstjüngere Version → (Unterschied)
Zur Navigation springen Zur Suche springen

EfficientNet: For scalability

EfficientNet use something called compound scaling. It scale the network’s width, depth, and resolution all together in a balanced way. Instead of just making the network deeper like ResNet, or wider only, EfficientNet try to find the best balance between them. Because of this, it can reach higher accuracy with less parameters and less FLOPs (Floating Point Operations).[1]

Comparison: EfficientNet-B0, which is the base model, can reach almost the same accuracy as ResNet-50 but with about 1/5 of the parameters and FLOPs. This make it a very strong choice for modern research, where efficiency is important not only raw accuracy alone.[1]

References

  1. 1,0 1,1 Joshua, Chidiebere & Kotsis, Konstantinos & Ghosh, Sourangshu. (2025). Comparative Evaluation of ResNet, EfficientNet, and MobileNet for Accurate Classification of Babylonian Sexagesimal Numerals.