Setup

Make sure to have Python3 installed. If you would like to use a gpu, install CUDA and CUDNN. Next, run:

pip install -U Cython==0.29.15 numpy==1.17.4
pip install -r requirements.txt

Then install either requirements-cpu.txt or requirements-gpu.txt depending on your device.

Data Generation

To create data to train an object detector on, run:

PYTHONPATH=$(pwd) data_generation/create_detection_data.py 

To create data to train a classifier, run:

PYTHONPATH=$(pwd) data_generation/create_clf_data.py 

Edit the data_generation/config.yaml to adjust the amount of data to create.

Classifier Training

See example model configurations in models/configs for inspiration on a model architecture. Check out torchvision for all the possibilities. An example training command is:

PYTHONPATH=$(pwd) train/train_clf.py \
    --model_config configs/resnet18.yaml 

Detector Training

See example model configurations in models/configs for inspiration on a model architecture.

PYTHONPATH=$(pwd) train/train_det.py \
    --model_config configs/vovnet-det.yaml

Inference on an Image

Inference on an image will utilize both the classifier and object detector model. The inference script can be run as follows:

PYTHONPATH=$(pwd) inference/find_targets.py \
    --image_path /path/to/image \
    --visualization_dir /path/to/save

The command above will visualize and save the models' predictions. See inference/find_targets.py for the full list of available arguments.

Testing

Testing is performed by running tox at the repo root. The test in this repo consist of python unittests and doctests.

Other

Updating submodules: git submodule update --remote --merge

TODO

• Devise a way to be able to create detection, classification, and feature data in separate independently.

• Finalize training pipelines.

• Finish inference pipeline.

• Work on Box Dev integration if platform available.

• Weights & Biases integration once project is granted.

• GitHub Actions CI and unittests