Bag of Tricks and A Strong Baseline for Deep Person Re-identification. CVPRW2019, Oral.
A Strong Baseline and Batch Normalization Neck for Deep Person Re-identification. IEEE Transactions on Multimedia (Accepted).
[Journal Version(TMM)] [PDF] [Slides] [Poster]
We are very grateful for your contribution to our project and hope that this project can help your research or work.
The codes are expanded on a ReID-baseline , which is open sourced by our co-first author Xingyu Liao.
Another re-implement is developed by python2.7 and pytorch0.4. [link]
A tiny repo with simple re-implement. [link]
Our baseline also achieves great performance on Vehicle ReID task! [link]
With Ranked List loss(CVPR2019)[link], our baseline can achieve better performance. [link]
@InProceedings{Luo_2019_CVPR_Workshops,
author = {Luo, Hao and Gu, Youzhi and Liao, Xingyu and Lai, Shenqi and Jiang, Wei},
title = {Bag of Tricks and a Strong Baseline for Deep Person Re-Identification},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}
@ARTICLE{Luo_2019_Strong_TMM,
author={H. {Luo} and W. {Jiang} and Y. {Gu} and F. {Liu} and X. {Liao} and S. {Lai} and J. {Gu}},
journal={IEEE Transactions on Multimedia},
title={A Strong Baseline and Batch Normalization Neck for Deep Person Re-identification},
year={2019},
pages={1-1},
doi={10.1109/TMM.2019.2958756},
ISSN={1941-0077},
}
We support
- easy dataset preparation
- end-to-end training and evaluation
- high modular management
- speed up inference [link]
- support multi-gpus training [link]
Bag of tricks
- Warm up learning rate
- Random erasing augmentation
- Label smoothing
- Last stride
- BNNeck
- Center loss
In the future, we will
- [] support more datasets
- [] support more models
- [] explore more tricks
Model | Market1501 | DukeMTMC-reID |
---|---|---|
Standard baseline | 87.7 (74.0) | 79.7 (63.8) |
+Warmup | 88.7 (75.2) | 80.6(65.1) |
+Random erasing augmentation | 91.3 (79.3) | 81.5 (68.3) |
+Label smoothing | 91.4 (80.3) | 82.4 (69.3) |
+Last stride=1 | 92.0 (81.7) | 82.6 (70.6) |
+BNNeck | 94.1 (85.7) | 86.2 (75.9) |
+Center loss | 94.5 (85.9) | 86.4 (76.4) |
+Reranking | 95.4 (94.2) | 90.3 (89.1) |
Backbone | Market1501 | DukeMTMC-reID |
---|---|---|
ResNet18 | 91.7 (77.8) | 82.5 (68.8) |
ResNet34 | 92.7 (82.7) | 86.4(73.6) |
ResNet50 | 94.5 (85.9) | 86.4 (76.4) |
ResNet101 | 94.5 (87.1) | 87.6 (77.6) |
ResNet152 | 80.9 (59.0) | 87.5 (78.0) |
SeResNet50 | 94.4 (86.3) | 86.4 (76.5) |
SeResNet101 | 94.6 (87.3) | 87.5 (78.0) |
SeResNeXt50 | 94.9 (87.6) | 88.0 (78.3) |
SeResNeXt101 | 95.0 (88.0) | 88.4 (79.0) |
IBN-Net50-a | 95.0 (88.2) | 90.1 (79.1) |
The designed architecture follows this guide PyTorch-Project-Template, you can check each folder's purpose by yourself.
-
cd
to folder where you want to download this repo -
Run
git clone https://github.com/michuanhaohao/reid-strong-baseline.git
-
Install dependencies:
- pytorch>=0.4
- torchvision
- ignite=0.1.2 (Note: V0.2.0 may result in an error)
- yacs
-
Prepare dataset
Create a directory to store reid datasets under this repo or outside this repo. Remember to set your path to the root of the dataset in
config/defaults.py
for all training and testing or set in every single config file inconfigs/
or set in every single command.You can create a directory to store reid datasets under this repo via
cd reid-strong-baseline mkdir data
(1)Market1501
- Download dataset to
data/
from http://www.liangzheng.org/Project/project_reid.html - Extract dataset and rename to
market1501
. The data structure would like:
data market1501 # this folder contains 6 files. bounding_box_test/ bounding_box_train/ ......
(2)DukeMTMC-reID
- Download dataset to
data/
from https://github.com/layumi/DukeMTMC-reID_evaluation#download-dataset - Extract dataset and rename to
dukemtmc-reid
. The data structure would like:
data dukemtmc-reid DukeMTMC-reID # this folder contains 8 files. bounding_box_test/ bounding_box_train/ ......
- Download dataset to
-
Prepare pretrained model if you don't have
(1)ResNet
from torchvision import models models.resnet50(pretrained=True)
(2)Senet
import torch.utils.model_zoo as model_zoo model_zoo.load_url('the pth you want to download (specific urls are listed in ./modeling/backbones/senet.py)')
Then it will automatically download model in
~/.torch/models/
, you should set this path inconfig/defaults.py
for all training or set in every single training config file inconfigs/
or set in every single command.(3)ResNet_IBN_a
You can download the ImageNet pre-trained weights from here [link]
(4)Load your self-trained model If you want to continue your train process based on your self-trained model, you can change the configuration
PRETRAIN_CHOICE
from 'imagenet' to 'self' and set thePRETRAIN_PATH
to your self-trained model. We offerExperiment-pretrain_choice-all_tricks-tri_center-market.sh
as an example. -
If you want to know the detailed configurations and their meaning, please refer to
config/defaults.py
. If you want to set your own parameters, you can follow our method: create a new yml file, then set your own parameters. Add--config_file='configs/your yml file'
int the commands described below, then our code will merge your configuration. automatically.
You can run these commands in .sh
files for training different datasets of differernt loss. You can also directly run code sh *.sh
to run our demo after your custom modification.
- Market1501, cross entropy loss + triplet loss
python3 tools/train.py --config_file='configs/softmax_triplet.yml' MODEL.DEVICE_ID "('your device id')" DATASETS.NAMES "('market1501')" OUTPUT_DIR "('your path to save checkpoints and logs')"
- DukeMTMC-reID, cross entropy loss + triplet loss + center loss
python3 tools/train.py --config_file='configs/softmax_triplet_with_center.yml' MODEL.DEVICE_ID "('your device id')" DATASETS.NAMES "('dukemtmc')" OUTPUT_DIR "('your path to save checkpoints and logs')"
You can test your model's performance directly by running these commands in .sh
files after your custom modification. You can also change the configuration to determine which feature of BNNeck is used and whether the feature is normalized (equivalent to use Cosine distance or Euclidean distance) for testing.
Please replace the data path of the model and set the PRETRAIN_CHOICE
as 'self' to avoid time consuming on loading ImageNet pretrained model.
- Test with Euclidean distance using feature before BN without re-ranking,.
python3 tools/test.py --config_file='configs/softmax_triplet_with_center.yml' MODEL.DEVICE_ID "('your device id')" DATASETS.NAMES "('market1501')" TEST.NECK_FEAT "('before')" TEST.FEAT_NORM "('no')" MODEL.PRETRAIN_CHOICE "('self')" TEST.WEIGHT "('your path to trained checkpoints')"
- Test with Cosine distance using feature after BN without re-ranking,.
python3 tools/test.py --config_file='configs/softmax_triplet_with_center.yml' MODEL.DEVICE_ID "('your device id')" DATASETS.NAMES "('market1501')" TEST.NECK_FEAT "('after')" TEST.FEAT_NORM "('yes')" MODEL.PRETRAIN_CHOICE "('self')" TEST.WEIGHT "('your path to trained checkpoints')"
- Test with Cosine distance using feature after BN with re-ranking
python3 tools/test.py --config_file='configs/softmax_triplet_with_center.yml' MODEL.DEVICE_ID "('your device id')" DATASETS.NAMES "('dukemtmc')" TEST.NECK_FEAT "('after')" TEST.FEAT_NORM "('yes')" MODEL.PRETRAIN_CHOICE "('self')" TEST.RE_RANKING "('yes')" TEST.WEIGHT "('your path to trained checkpoints')"