Encord-Active MaskRCNN Example

Important Note: This project is designed to be used for the Encord-Active data centric challenge, please use the original project for training Encord-Active projects

This example implements an end-to-end training and evaluation procedure for Mask R-CNN in PyTorch using Encord Active.

• The target project should follow the COCO format. You can use Encord-Active's export functionality to create COCO annotations.
• Trained model can be evaluated on any dataset using the Encord-Active.

Installation

Create a new conda virtual environment using the following command

# inside the project directory
conda env create -f environment.yml

Verify that the new environment is installed correctly:

conda env list

You should see the encord-maskrcnn example in the list. Simply activate it with the following command:

conda activate encord-maskrcnn

Training

1. You can use Encord-Active's Actions tab to create COCO annotations.
2. Create a config.ini file by looking at the example_config.ini
3. You can resize the images and corresponding annotations via utils/downscale_dataset.py.
4. For the training, the only required fields are [DATA], [LOGGING], and [TRAIN] sections.
5. Activate the environment and run python train.py
6. You can track the progress of the training on wandb platform.

Importing Encord-Active predictions

1. Get the wandb ID from the experiment that you want to use for inference
2. Fill the [INFERENCE] section of the config.ini file
3. Run python generate_ea_predictions.py to generate a pickle file.
4. Run the following command to convert pickle file into Encord Active predictions.

encord-active import predictions /path/to/predictions.pkl -t /path/to/project

5. Open the app. Now you can see the model performance on the Model Quality tab.