- Image Data: The primary input is the image or set of images that you want to segment. These images are typically in formats like JPEG, PNG, etc.
- Ground Truth Labels: These are labeled images where each pixel's value corresponds to a class label. These are used for training the model.
- Pre-trained Network: You can use a pre-trained network like SegNet, U-Net, or DeepLab for transfer learning.
- Training Parameters: Parameters like learning rate, batch size, number of epochs, etc.
- Segmented Image: The result is an image where each pixel is labeled with its corresponding class.
- Trained Model: If you train a new model, you'll get a trained network that can be used for future segmentation tasks.
- Metrics: Performance metrics like accuracy, IoU (Intersection over Union), etc., to evaluate the model's performance.