MMVRAC2024-Track12-Solution

Welcome to the MMVRAC2024 Track12 solution repository. This repository contains the implementation of our approach for the Track12 challenge. You can also obtain the code from Google Drive: Link.

Dataset

To get started with the solution, you'll need access to the UAV-Human dataset. Please refer to the following link to obtain the dataset and read the instructions for setup: UAV-Human Dataset

Quick Start

To replicate our results or use the solution as a starting point for your experiments, follow these steps:

1. Prepare dataset

Create a directory to store reid datasets under this repo.

cd UAV_ReID
mkdir toDataset

• Download the pre-trained ResNet model file resnet152-b121ed2d.pth from the link and place it in the pre-train directory within your project.
• Update the configs/AGW_uavhuman.yml file by setting the PRETRAIN_PATH field to the path where your resnet152-b121ed2d.pth is located. It should look something like this: PRETRAIN_PATH: 'home/UAV_ReID/pre-train/resnet152-b121ed2d.pth'
• Download UAV-Human Dataset to toDataset/
• Unzip the uavhuman.zip. The data structure would like:

toDataset
    uavhuman
        bounding_box_test/
        bounding_box_train/
        query/
        ......

2. Install dependencies

• pytorch=1.0.0
• torchvision=0.2.1
• pytorch-ignite=0.1.2
• yacs
• scipy=1.2.1
• h5py

3. Train

To train a AGW model with on UAV-Human with GPU device 0, run similarly:

python3 tools/main.py --config_file='configs/AGW_uavhuman.yml' MODEL.DEVICE_ID "('0')" DATASETS.NAMES "('uavhuman')" OUTPUT_DIR "('./log/uavhuman/Experiment-AGW-uavhuman')"

4. Test

To test a AGW model with on UAV-Human with weight file '/home/user/UAV_ReID/log/uavhuman/Experiment-AGW-uavhuman/resnet152_nl_model_120.pth',(You should to replace the path with your file path).

If you want to do a quick test, you can download the checkpoint and replace the path for TEST.WEIGHT with the path of the checkpoint you downloaded.

Run similarly:

python3 tools/main.py --config_file='configs/AGW_uavhuman.yml' MODEL.DEVICE_ID "('0')" DATASETS.NAMES "('uavhuman')"  MODEL.PRETRAIN_CHOICE "('self')" TEST.WEIGHT "('/home/user/UAV_ReID/log/uavhuman/Experiment-AGW-uavhuman/resnet152_nl_model_120.pth')" TEST.EVALUATE_ONLY "('on')" OUTPUT_DIR "('./log/Test')"

Acknowledgement

This repository is built based on AGW and UAV-Human.

Thanks for their excellent work.

Contact

If you have any questions, feel free to contact Linze or Jian Ding by email.