This repository is for Global Wheat Detection
that is Kaggle competition for detection of wheat segments in images from all over the world. (https://www.kaggle.com/c/global-wheat-detection)
We implemented a pipeline for object detecion with pytorch. You can run training and inference easily in your machines.
In our pipeline, the following parameters are currently available, as well as TensorBoard and mlflow for visualization.
Models
Optimizers
Schedulers
Augmentations
Post processing
Other utils
tensorboard --logdir ./output
in root)mlflow ui --port 5000
in root)Please see the requirements.txt. Run the following command to install. pip install -U -r requirements.txt
Specify your training configuration as a .json script, and then run the following command.
$ python train.py YOUR_CONFIG_PATH
You can use Single GPU. Multi GPU will be available soon!
This model is from https://github.com/rwightman/efficientdet-pytorch.
We used the following command to get the pretrained model (also from the above website)
$ wget https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d5-ef44aea8.pth
In order to use this, please specify the path to this weight in config.json. Example is given in ./sample_json/config_effdet.json.