Calculate Dice and IoU metrics per class

Create colored visualization of segmentation mask

Run inference on a single image Args: model: trained segmentation model image_path: path to input image ou

4-direction TTA: original + horizontal flip + vertical flip + both flips Average predictions for more robust results

Data augmentation for training

Match image files with their corresponding label JSON files

Test-time augmentation prediction

Convert numpy array to torch tensor

Create segmentation mask from polygon annotations

Create segmentation model

Run inference on all images in a directory

Parse LabelMe JSON annotation file

Resize mask back to original size

4-direction TTA: average predictions from original and flipped versions

Load and preprocess image

Train for one epoch

Validate model

Validate with TTA on specified indices

Create overlay visualization Args: image: PIL Image or numpy array mask: segmentation mask

Visualize all samples in dataset

Visualize single prediction

Visualize original image, mask, and overlay

Create overlay of image and segmentation mask

Load model weights from checkpoint

Load and preprocess image

Load a new segmentation model

Predict segmentation mask for an image Args: image: PIL Image or numpy array (H, W, 3) input_size: Targ

Visualize segmentation results