sekilab / vehiclereidentificationdataset

Vehicle Re-Identification (ReID) dataset contains over 55,000 images for training and validation of the vehicle re-identification model

Vehicle Re-Identification (ReID) Dataset

Sample images from the CARLA ReID dataset, resized to the same dimension for visualization. Figure on the left shows the variation of vehicle models and colors in the CARLA ReID dataset. CARLA ReID dataset contains four classes of vehicles such as car, truck, motorcycle, and bicycle. Figure on the right shows the camera angle variation for observing the same vehicle. Each vehicle is observed by 85 cameras located at varying distances surrounding the target vehicle.

• The file names contain the information on vehicle id and camera id separated by an underscore, necessary to train ReID model.

  • If the filename is 20220711212617_24_9.jpg, then 20220711212617 is the datetime, 24 is the camera id and 9 represents the vehicle id. These two information is enough for training a ReID model

Download links

The download URL of the VeRi dataset is shown here. Please note that training, gallery and query images are all inside the same zip file.

Download VeRi dataset (CARLA)

Training example and Installation

• Please check training_vehicle_reid.ipynb Jupyter notebook to train the model or understand the training process, in general.

Installation

• Make sure that you have conda installed. Let's create a new virtual environment carla_reid with Python verison 3.7 by executing the following command:

conda create -n carla_reid python=3.7 pip pandas  -y

• The above command, however, will not install torchreid. Also, pip package does not support latest build of torchreid and it needs to built from the source. I have provided bash script to simplify this task

• First of all, activate the environment created before by executing the following command:

conda activate carla_reid

• Make sure that the environment is activated and you are in the root directory of this repository. It should show (carla_reid) on the left of your username in the terminal. Run the following command:

bash install_packages_reid.sh

• The above command will install torchreid 1.4.0 along with dependencies such as torch, torchvision, etc. You are all set to run the Jupyter notebook.

Citations

If you find our dataset useful and use it in your research, please give us a star ⭐ and cite the following article:

@INPROCEEDINGS{10020814,
  author={Kumar, Ashutosh and Kashiyama, Takehiro and Maeda, Hiroya and Zhang, Fan and Omata, Hiroshi and Sekimoto, Yoshihide},
  booktitle={2022 IEEE International Conference on Big Data (Big Data)}, 
  title={Vehicle re-identification and trajectory reconstruction using multiple moving cameras in the CARLA driving simulator}, 
  year={2022},
  volume={},
  number={},
  pages={1858-1865},
  doi={10.1109/BigData55660.2022.10020814}}

@article{kumar2022vehicle,
  title={Vehicle re-identification and trajectory reconstruction using multiple moving cameras in the CARLA driving simulator},
  author={Kumar, Ashutosh and Kashiyama, Takehiro and Maeda, Hiroya and Zhang, Fan and Omata, Hiroshi and Sekimoto, Yoshihide},
  year={2022},
  publisher={TechRxiv}
}

vehiclereidentificationdataset's Issues

datamanager exits with KeyError: 3

Thanks for the dataset, it's a good intro data into getting started with Re-ID.
However, i've encountered an issue while following the instructions xD

# You can change these hyperparameters as you wish
datamanager = torchreid.data.ImageDataManager(

    root='',
    sources='synthtetic_veri_reid',
    height = 256,
    width = 256,
    transforms =['random_flip', 'random_crop'])

The above block raises a KeyError: 3 from class VeRiDataset car_id = label_dict[int(image.split('.jpg')[0].split('_')[-1])]

Is it that the dictionary isn't getting assigned the required key from that split operation ?

Also, the torchreid instructions don't have the root empty as per the custom dataset example

torchreid module wasn't in the condamdx_reid.yml for which i've installed torchreid==0.2.5 through pip

I suppose the error is from the split logic or is there anything else that might be causing it ?
path_to_dataset is pointing to the directory that contains the three data folders of gallery, query and train.

attaching the full trace-back below for reference:
##############################################################################

Building train transforms ...

• resize to 256x256
• random flip
• random crop (enlarge to 288x288 and crop 256x256)
• to torch tensor of range [0, 1]
• normalization (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
  Building test transforms ...
• resize to 256x256
• to torch tensor of range [0, 1]
• normalization (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
  => Loading train (source) dataset

---

KeyError Traceback (most recent call last)
Input In [7], in <cell line: 2>()
1 # You can change these hyperparameters as you wish
----> 2 datamanager = torchreid.data.ImageDataManager(
3 root='reid-data',
4 sources='synthtetic_veri_reid',
5 height = 256,
6 width = 256,
7 transforms =['random_flip', 'random_crop'])

File ~/anaconda3/envs/mdx_reid/lib/python3.9/site-packages/torchreid/data/datamanager.py:195, in ImageDataManager.init(self, root, sources, targets, height, width, transforms, k_tfm, norm_mean, norm_std, use_gpu, split_id, combineall, load_train_targets, batch_size_train, batch_size_test, workers, num_instances, num_cams, num_datasets, train_sampler, train_sampler_t, cuhk03_labeled, cuhk03_classic_split, market1501_500k)
193 trainset = []
194 for name in self.sources:
--> 195 trainset_ = init_image_dataset(
196 name,
197 transform=self.transform_tr,
198 k_tfm=k_tfm,
199 mode='train',
200 combineall=combineall,
201 root=root,
202 split_id=split_id,
203 cuhk03_labeled=cuhk03_labeled,
204 cuhk03_classic_split=cuhk03_classic_split,
205 market1501_500k=market1501_500k
206 )
207 trainset.append(trainset_)
208 trainset = sum(trainset)

File ~/anaconda3/envs/mdx_reid/lib/python3.9/site-packages/torchreid/data/datasets/init.py:42, in init_image_dataset(name, **kwargs)
37 if name not in avai_datasets:
38 raise ValueError(
39 'Invalid dataset name. Received "{}", '
40 'but expected to be one of {}'.format(name, avai_datasets)
41 )
---> 42 return __image_datasetsname

Input In [4], in VeRiDataset.init(self, root, **kwargs)
46 for image in IMAGES:
47 cam_id = int(image.split('.jpg')[0].split('')[-2])
---> 48 car_id = label_dict[int(image.split('.jpg')[0].split('')[-1])]
49 query_array.append((FOLDER_PATH + dataset_type + "/" + image, car_id, cam_id))
52 train = training_array

KeyError: 3

##############################################################################

Thank you !