Skip to main content

Table 1 Methods we validated in the splits of training data in Caltech Pedestrian Detection Benchmark [8], sorted by validation error rates in descending order (corresponding to Fig. 3)

From: Pedestrian detection with motion features via two-stream ConvNets

Method

Spatial feature

Temporal feature

Temporal pre-training

Temporal input

ACF [19]

Hand-crafted

-

-

-

CCF [16]

VGGNet-16

-

-

-

CCF + VGG-SF

VGGNet -16

VGGNet-16

ILSVRC2012

SF

CCF + SDt

VGGNet-16

SDt itself

-

SDt

CCF + VGG-SDt

VGGNet-16

VGGNet-16

ILSVRC2012

SDt

CCF + UCFAlex-SDt

VGGNet-16

UCFAlex

UCF-101

SDt