FastREID性能评测

FastReID Model Zoo and Baselines

Introduction

This file documents collection of baselines trained with fastreid. All numbers were obtained with 1 NVIDIA P40 GPU.
The software in use were PyTorch 1.4, CUDA 10.1.

In addition to these official baseline models, you can find more models in projects/.

How to Read the Tables

• The “Name” column contains a link to the config file.
  Running tools/train_net.py with this config file and 1 GPU will reproduce the model.
• The model id column is provided for ease of reference. To check downloaded file integrity, any model on this page contains tis md5 prefix in its file name.
• Training curves and other statistics can be found in metrics for each model.

Common Settings for all Person reid models

BoT:

Bag of Tricks and A Strong Baseline for Deep Person Re-identification. CVPRW2019, Oral.

AGW:

ReID-Survey with a Powerful AGW Baseline.

MGN:

Learning Discriminative Features with Multiple Granularities for Person Re-Identification

SBS:

stronger baseline on top of BoT:

Bag of Freebies(BoF):

1. Circle loss
2. Freeze backbone training
3. Cutout data augmentation & Auto Augmentation
4. Cosine annealing learning rate decay
5. Soft margin triplet loss

Bag of Specials(BoS):

1. Non-local block
2. GeM pooling

Market1501 Baselines

BoT:

Method	Pretrained	Rank@1	mAP	mINP	download
BoT(R50)	ImageNet	94.4%	86.1%	59.4%	-
BoT(R50-ibn)	ImageNet	94.9%	87.6%	64.1%	-
BoT(S50)	ImageNet	95.1%	88.5%	66.0%	-
BoT(R101-ibn)	ImageNet	95.4%	88.9%	67.4%	-

AGW:

Method	Pretrained	Rank@1	mAP	mINP	download
AGW(R50)	ImageNet	95.3%	88.2%	66.3%	-
AGW(R50-ibn)	ImageNet	95.1%	88.7%	67.1%	-
AGW(S50)	ImageNet	94.7%	87.1%	62.2%	-
AGW(R101-ibn)	ImageNet	95.5%	89.5%	69.5%	-

SBS:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50)	ImageNet	95.4%	88.2%	64.8%	-
SBS(R50-ibn)	ImageNet	95.7%	89.3%	67.5%	-
SBS(S50)	ImageNet	95.0%	87.0%	60.6%	-
SBS(R101-ibn)	ImageNet	96.3%	90.3%	70.0%	-

MGN:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50-ibn)	ImageNet	95.8%	89.7%	67.0%	-

DukeMTMC Baseline

BoT:

Method	Pretrained	Rank@1	mAP	mINP	download
BoT(R50)	ImageNet	87.1%	76.9%	41.6%	-
BoT(R50-ibn)	ImageNet	89.6%	79.1%	44.4%	-
BoT(S50)	ImageNet	87.8%	77.7%	39.6%	-
BoT(R101-ibn)	ImageNet	91.1%	81.3%	47.7%	-

AGW:

Method	Pretrained	Rank@1	mAP	mINP	download
AGW(R50)	ImageNet	89.0%	79.9%	46.3%	-
AGW(R50-ibn)	ImageNet	89.8%	80.7%	47.7%	-
AGW(S50)	ImageNet	89.9%	79.7%	44.2%	-
AGW(R101-ibn)	ImageNet	91.4%	82.1%	50.2%	-

SBS:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50)	ImageNet	89.6%	79.8%	44.6%	-
SBS(R50-ibn)	ImageNet	91.3%	81.6%	47.6%	-
SBS(S50)	ImageNet	90.5%	79.1%	42.7%	-
SBS(R101-ibn)	ImageNet	92.4%	83.2%	49.7%	-

MGN:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50-ibn)	ImageNet	91.6%	82.1%	46.7%	-

MSMT17 Baseline

BoT:

Method	Pretrained	Rank@1	mAP	mINP	download
BoT(R50)	ImageNet	72.3%	48.3%	9.7%	-
BoT(R50-ibn)	ImageNet	77.0%	54.4%	12.5%	-
BoT(S50)	ImageNet	80.4%	59.2%	15.9%	-
BoT(R101-ibn)	ImageNet	79.0%	57.5%	14.6%	-

AGW:

Method	Pretrained	Rank@1	mAP	mINP	download
AGW(R50)	ImageNet	76.7%	53.6%	12.2%	-
AGW(R50-ibn)	ImageNet	79.3%	57.5%	14.3%	-
AGW(S50)	ImageNet	77.3%	54.7%	12.6%	-
AGW(R101-ibn)	ImageNet	80.8%	60.2%	16.5%	-

SBS:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50)	ImageNet	83.3%	59.9%	14.6%	-
SBS(R50-ibn)	ImageNet	84.0%	61.2%	15.5%	-
SBS(S50)	ImageNet	82.6%	58.2%	13.2%	-
SBS(R101-ibn)	ImageNet	85.1%	63.3%	16.6%	-

MGN:

Method	Pretrained	Rank@1	mAP	mINP	download
SBS(R50-ibn)	ImageNet	85.1%	65.4%	18.4%	-

VeRi Baseline

SBS:

Method	Pretrained	Rank@1	mAP	mINP	download
BoT(R50-ibn)	ImageNet	97.0%	81.9%	46.3%	-

VehicleID Baseline

BoT:
Test protocol: 10-fold cross-validation; trained on 4 NVIDIA P40 GPU.

Method	Pretrained	Testset size						download
		Small		Medium		Large
		Rank@1	Rank@5	Rank@1	Rank@5	Rank@1	Rank@5
BoT(R50-ibn)	ImageNet	86.6%	97.9%	82.9%	96.0%	80.6%	93.9%	-

VERI-Wild Baseline

BoT:
Test protocol: Trained on 4 NVIDIA P40 GPU.

Method	Pretrained	Testset size									download
		Small			Medium			Large
		Rank@1	mAP	mINP	Rank@1	mAP	mINP	Rank@1	mAP	mINP
BoT(R50-ibn)	ImageNet	96.4%	87.7%	69.2%	95.1%	83.5%	61.2%	92.5%	77.3%	49.8%	-