🚀 GRAVITY NET

GravityNet is a novel one-stage end-to-end detector specifically designed to detect small lesions in medical images.

GravityNet framework is available on GitHub

📥 INSTALLATION

Install the package running:

pip install gravitynet

Import the package as:

import gravitynet

📁 PACKAGE ORGANIZATION

GravityNet is structured into three main modules:

🔧 gravity_points_config

Generates the configuration of gravity points.

gravity_points, num_gravity_points, num_gravity_points_feature_map = gravity_points_config(config,
                                                                                           image_shape)

PARAMETERS
config: type of configuration:
grid: grid-base configuration for gravity-points (e.g., grid-10)
dice: dice-base configuration for gravity-points (e.g., dice-1)

image_shape: the shape of the image (H x W)

RETURNS
gravity_points: configuration of the gravity points on the image
num_gravity_points: number of gravity points generated
num_gravity_points_feature_map: number of gravity points generated per feature map

🌐 GravityNet

Define the GravityNet model.

net = GravityNet(backbone,
                 pretrained,
                 num_gravity_points_feature_map)

PARAMETERS
backbone: backbone model (e.g., ResNet)
pretrained: pretrained option
num_gravity_points_feature_map: number of gravity points generated per feature map

RETURNS
net: GravityNet model

📉 GravityLoss

Define the GravityLoss function used for training.

criterion = GravityLoss(config,
                        hook,
                        num_gravity_points_feature_map,
                        device)

PARAMETERS
config: type of configuration (e.g., grid-10)
hook: hooking distance (e.g., 10)
num_gravity_points_feature_map: number of gravity points generated per feature map
device: device (e.g., cuda)

RETURNS
criterion: GravityLoss criterion