3D ResNet Paper Crosses 3,000 Citations Eight Years After CVPR 2018

HirokatuKataoka · x · 2026-09-11

The CVPR 2018 paper "Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet?" (with Kensho Hara and Yutaka Satoh) has crossed 3,000 citations.

The work asked whether a large-scale video dataset could play the role ImageNet played for 2D CNNs: it showed the Kinetics dataset enabled training very deep 3D CNNs of its era (including 3D ResNet-152) and established 3D ResNets as strong architectures for video representation learning.

Paper and code remain publicly available and widely used eight years later.

Original post →

More from Multimodal

Multimodal channel →