Modification of cosine_metric_learning from https://github.com/nwojke/cosine_metric_learning
This repository is heavily borrowed from Deep Cosine Metric Learning for Person Re-identification(https://github.com/nwojke/cosine_metric_learning).
The original github project is trained on Mars and Market1501 pedestrian dataset and aimed to extract the 128 dimensional feature of the pedestrain to achieve person re-identification.
There are two files added in this repository to train on the VeRi dataset:
- train_veri.py: The training code for vehicle re-id.
- datasets/veri.py: The dataset preprocessing for VeRi dataset.
This vehicle re-id is mainly used for deep_sort tracker.
The VeRi dataset contains over 50,000 images of 776 vehicles captured by 20 cameras covering an 1.0 km^2 area in 24 hours, which makes the dataset scalable enough for vehicle Re-Id and other related research. see this page for more information(https://github.com/VehicleReId/VeRidataset).
If you need to use this dataset, please contact the author of VeRi.
The following description assumes you have downloaded the VeRi dataset to ./VeRi. The following command starts training using the cosine-softmax classifier described in this paper(https://elib.dlr.de/116408/):
python train_veri.py \
--dataset_dir=./VeRi/ \
--loss_mode=cosine-softmax \
--log_dir=./output/veri/ \
--run_id=cosine-softmaxThis will create a directory ./output/veri/cosine-softmax where TensorFlow checkpoints are stored and which can be monitored using tensorboard:
tensorboard --logdir ./output/veri/cosine-softmax --port 6006Sometimes there would be some bugs with the tensorboard and the site can not be displaied. You can try:
tensorboard --logdir ./output/veri/cosine-softmax --port 8080To export your trained model for use with the deep_sort tracker, run the following command:
python train_veri.py --mode=freeze --restore_path=PATH_TO_CHECKPOINT