olivierlaflamme / datacvchallenge-2024-forest-classifier

Repository for our code for DataCV Challenge in CVPR 2024

DataCV2024

This is the development kit repository for the 2nd CVPR DataCV Challenge. Here you can find details on how to download datasets.

Running Environment for Classifier Guided Cluster Density Reduction (CCDR)

To manage the python code, it is recommended to install Anaconda.

For creating environment,

conda create -n ccdr python=3.10 -y
conda activate ccdr

Besides, you will need to install pytorch 2.0.0, please modify cuda version according to the version installed in the system.

pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118

The install of mmcv library.

pip install -U openmim
mim install mmcv-full==1.7.2
pip install yapf==0.40.1

The install of mmdetction.

cd task_model/mmdetection/
pip install -v -e . 

Additionally, other required libraries can be installed with the following command:

pip install -r  requirements.txt

The installation of pycococreator Method 1:

pip install git+git://github.com/waspinator/pycococreator.git@0.2.0

Method 2:

1. install cython

pip install Cython

2. Download the repo, https://github.com/waspinator/pycococreator
3. Extract the repo
4. Go to the repo directory and run the command:

python setup.py install

The installation of jax

pip install --upgrade "jax[cuda11_pip]==0.4.20" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

Set up Scenic git clone the scenic and run the command:

cd DataCV2024_comp
git clone https://github.com/google-research/scenic.git
cd /path/to/scenic
python setup.py install

Running example

For running such a process, when region100 is used as the target, we can search a training set with 8000 images using the command below:

python trainingset_search_detection_vehicle.py --target 'region100' \
--select_method 'CCDR' --c_num 50 \
--result_dir 'main_results/sample_data_detection_vehicle_region100/' \
--n_num 8000 \
--output_data 'CCDR_region100_vehicle_8000_c_num50.json' 

To use the classifier to score selected candidates, see classifier/README.md for details.

Reproducing LB results

The trained weights can be downloaded from(LB weights for trained model To perform inference

python tools/test.py DataCV2024-main/task_model/mmdetection/workdirs/retinanet_r50_fpn_1x_custom_tss_car.py DataCV2024-main/task_model/mmdetection/workdirs/epoch_12.pth --eval bbox

To produce our leaderboard submission for the competition:

python tools/test.py  DataCV2024-main/task_model/configs_tss/retinanet/retinanet_r50_fpn_1x_custom_tss_car.py DataCV2024-main/task_model/mmdetection/workdirs/epoch_12.pth  --format-only --options "jsonfile_prefix=./"