Local Information Matters: A Rethink of Crowd Counting

Benchmark Model Rank Results
crowd-counting-on-shanghaitech-aLIMM + ConvNeXt-T–MAE: 50.8MSE: 84.2
crowd-counting-on-shanghaitech-aLIMM + ResNet-50–MAE: 59.2MSE: 89.0
crowd-counting-on-shanghaitech-aLIMM + Swin-T–MAE: 52.2MSE: 84.6
crowd-counting-on-shanghaitech-bLIMM + ConvNeXt-T–MAE: 6.5
crowd-counting-on-shanghaitech-bLIMM + ResNet-50–MAE: 6.9
crowd-counting-on-shanghaitech-bLIMM + Swin-T–MAE: 6.8
crowd-counting-on-ucf-qnrfLIMM + ConvNeXt-T–MAE: 76.4
crowd-counting-on-ucf-qnrfLIMM + ResNet-50–MAE: 88.9
crowd-counting-on-ucf-qnrfLIMM + Swin-T–MAE: 78.5