mask-rcnn-documents 1.0.post20230911134841

No project description provided

Train and test a MASK-RCNN model

Tools to train and test a Mask-RCNN model using PyTorch.

He, Kaiming, Georgia Gkioxari, Piotr Dollár, et Ross Girshick. « Mask R-CNN ».
arXiv, 24 janvier 2018. http://arxiv.org/abs/1703.06870.

Installation

The scripts should be installed in your environment using pip by including the following line in your requirements.txt file.

git+ssh://[e-mail]/acadiie/document_layout_analysis/mask-rcnn.git#egg=mask-rcnn

Usage

train_mask_rcnn

usage: train_mask_rcnn [-h] --data-path DATA_PATH --experiment-name EXPERIMENT_NAME --epochs EPOCHS [--batch-size BATCH_SIZE] [--device DEVICE] [--log-path LOG_PATH] [--mlflow-endpoint MLFLOW_ENDPOINT]

Train a Mask-RCNN model based on ResNet-50

options:
  -h, --help            show this help message and exit
  --data-path DATA_PATH
                        Path to the directory where are stored the images and labels. In this directory, there should be a train and a val subdirectory, each containing another images and labels subdirectories.
  --experiment-name EXPERIMENT_NAME
                        Name of the experiment
  --epochs EPOCHS, -e EPOCHS
                        Number of epochs
  --batch-size BATCH_SIZE, -bs BATCH_SIZE
                        Batch size for the model
  --device DEVICE, -d DEVICE
                        Device on which to run the neural network
  --log-path LOG_PATH   Where to log the model steps
  --mlflow-endpoint MLFLOW_ENDPOINT
                        URL to a MLFlow tracking system

test_mask_rcnn

usage: Test Mask-RCNN model. [-h] --data-path DATA_PATH --model-checkpoint MODEL_CHECKPOINT [--threshold THRESHOLD]

options:
  -h, --help            show this help message and exit
  --data-path DATA_PATH
                        Path to the directory where are stored the images and labels. In this directory, there should be a test directory containing another images and labels subdirectories.
  --model-checkpoint MODEL_CHECKPOINT
                        Path to the model checkpoint to test
  --threshold THRESHOLD
                        Confidence threshold in the model to export the masks.