ray/COCO-Stuff-Segmentation - COCO Stuff 2017 Image Segmentation

Train Image Segmentation Models In Your Browser

Train an image segmentation model using the U-Net architecture. Load your image and segmentation masks to begin training.

Want to resume from a project you saved to your device?

Load project.zip

Load Images and Masks

Load your image and segmentation mask folders separately. Ensure your image and segmentation masks are named identically or differ by a common suffix . Don't have a dataset? Try COCO-Stuff .

What are segmentation masks and why might they be entirely black?

Load images folder

Load masks folder

1 / 1

Complete image-mask pairs: 0

Model Setup

Choose the type of model, its width and depth, and the size of your images

Select model type

ResUNet

Incorporates residual connections which transform the U-Net into a ResUNet.

U-Net

The classic U-Net introduced in the U-Net: Convolutional Networks for Biomedical Image Segmentation paper

Number of categories

Model depth

Feature detail level

Image height

Image width

Set Hyperparameters

Hyperparameters determine the type of model and how effectively it is trained. They are frequently adjusted to help the model reach its optimal state.

Training epochs

Batch size

Optimizer

Learning rate

Loss function

Validation split

Metrics (optional)

Train Model

Train the model until the dice loss is as close to zero as possible. A browser with strong WebGPU support (Chrome) is recommended for faster training. To prevent interruptions, keep this tab in the foreground. Browsers may throttle or sleep tabs that are not in view.

Epoch 1·Batch 2,000/2,000·00:43:31

Status: Model is idle

Predict

Run image segmentation directly in the browser. You can add color to distinguish segmented categories or just receive the raw outputs.

Load images

Drag and drop or paste images to segment