Baseline model (with bottleneck) for person ReID (using softmax and triplet loss).
We support
- easy dataset preparation
- end-to-end training and evaluation
- high modular management
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/L1aoXingyu/reid_baseline.git
-
Install dependencies:
- pytorch 1.0
- torchvision
- ignite
- yacs
-
Prepare dataset
Create a directory to store reid datasets under this repo via
cd reid_baseline mkdir data
- 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 bounding_box_test/ bounding_box_train/
- Download dataset to
-
Prepare pretrained model if you don't have
from torchvision import models models.resnet50(pretrained=True)
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/
.
Most of the configuration files that we provide, you can run this command for training market1501
python3 tools/train.py --config_file='configs/softmax.yml' DATASETS.NAMES "('market1501')"
You can also modify your cfg parameters as follow
python3 tools/train.py --config_file='configs/softmax.yml' INPUT.SIZE_TRAIN '(256, 128)' INPUT.SIZE_TEST '(256, 128)'
You can test your model's performance directly by running this command
python3 tools/test.py --config_file='configs/softmax.yml' TEST.WEIGHT '/save/trained_model/path'
network architecture
cfg | market1501 | cuhk03 | dukemtmc |
---|---|---|---|
softmax, size=(384, 128), batch_size=64 | 92.5 (79.4) | 60.4 (56.1) | 84.6 (68.1) |
softmax, size=(256, 128), batch_size=64 | 92.0 (80.4) | 60.5 (55.5) | 84.1(68.4) |
softmax_triplet, size=(384, 128), batch_size=128(32 id x 4 imgs) | 93.2 (82.5) | 64.2 (59.5) | 86.4 (73.1) |
softmax_triplet, size=(384, 128), batch_size=64(16 id x 4 imgs) | 93.8 (83.2) | 65.9 (61.4) | 86.2 (72.9) |