model_zoo.md

Introduction

• Please download the model in the table
• Please change the inference model path when you use these models
• Please follow the configurations in the experiments
• The results in parenthesis are proposed in the original papers

Method	Ped2	Avenue	Shanghai
STAE[1]	90.1(91.2)	75.2(77.1)	-
MemAE[2]	92.3(94.1)	82.9(83.3)	70.0 (71.2)
OCAE[5]	95.3(97.8)	90.0(90.4)	81.3 (84.9)
AnoPCN[7]	94.2(96.8)	87.1(86.2)	72.1 (73.6)
AMC[13]	97.4(96.2)	83.5 (86.9)	-
AnoPred[14]	93.9(96.4)	86.7 (85.1)	69.9 (72.8)

Note: We found that different GPU type may inference the results.

Details

Optical Flow

• Some methods using other optical flow methods to get the optical in video

• We choose to use the different optical flow methods implemented in PyTorch

  • FlowNet2
  • LiteFlowNet

• We use the pre-trained models in these methods, which can be download from their GitHub repo.

  Method	Optical Flow Method	Ped2	Avenue	Shanghai
  AMC	FlowNet2	97.4		-
  AMC	LiteFlowNet			-
  AnoPred	FlowNet2
  AnoPred	LiteFlowNet	89.5
  AnoPCN	FlowNet2
  AnoPCN	LiteFlowNet