OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Cct

CCT is an open-source Python implementation of Cross-Consistency Training, a semi-supervised method for semantic segmentation that encourages consistency across perturbed encoder representations. Its repository documents Pascal VOC training, extensions for weak image-level labels and multiple domains, checkpointing, and inference that saves pixel-level predictions as PNG files.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Cct — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use CCT when a research or model-development workflow calls for semi-supervised semantic segmentation and you can provide the dataset, Python environment and training configuration.

Who it is for

  • Computer-vision researchers studying semi-supervised semantic segmentation.
  • Developers who want to train or evaluate segmentation models with limited labeled examples.
  • Researchers exploring weak image-level labels or pixel-level labels across multiple domains.

Documented uses

  • Train a segmentation model using the CCT method and configurable labeled-example counts.
  • Generate pseudo-labels from image-level labels for the documented Pascal VOC workflow.
  • Resume training from a saved `.pth` checkpoint.
  • Apply a trained model to JPG images and save segmentation predictions as PNG files.
Before you choose

Requirements & limits

The implementation is available in a public GitHub repository under the MIT license.

  • This is a code repository that requires a local Python setup, dataset and configuration, not a hosted segmentation interface.
  • The documented base dataset is Pascal VOC 2012, augmented with SegmentationClassAug annotations.
  • The README lists PyTorch 1.1.0 or later as runnable, and reports testing with PyTorch 1.1.0 and CUDA 10.0.
  • Inference depends on both a pretrained checkpoint and the configuration used for training.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Train CCT on Pascal VOC

Prepare the documented Pascal VOC data and configure a CCT training run.

Before you start

  • A Python environment with PyTorch, torchvision, PIL, OpenCV and tqdm; dominate is listed among additional modules.
  • Pascal VOC 2012 and the additional SegmentationClassAug annotations.

Steps

  1. Download and extract Pascal VOC 2012, then add the SegmentationClassAug annotations to the VOC2012 directory.

  2. Install the repository requirements with the command below.

    Documented command
    pip install -r requirements.txt

  3. Set data_dir and other desired parameters in configs/config.json.

  4. Start training with the command below.

    Documented command
    python train.py --config configs/config.json

Expected outcome

Training logs, `.pth` checkpoints and HTML validation results are saved under `saved`.

Guide limitations

  • The repository describes Pascal VOC and its annotation augmentation for this workflow.
  • The README reports testing on Ubuntu 18.04.3, Python 3.7, PyTorch 1.1.0 and CUDA 10.0; it says recent PyTorch versions from 1.1.0 should be runnable.
Documentation 1 ↗
Source-based practical guide

Generate segmentation predictions

Run inference on JPG images using a pretrained or trained CCT model.

Before you start

  • A pretrained model checkpoint.
  • The configuration used for training.
  • A folder of JPG images.

Steps

  1. Obtain a pretrained model from the repository’s releases or use a trained checkpoint.

  2. Run the command below, substituting the paths to the configuration, model and image folder.

    Documented command
    python inference.py --config config.json --model best_model.pth --images images_folder

Expected outcome

The README says predictions are saved as PNG images in `outputs`.

Guide limitations

  • Inference requires a model and its training configuration.
  • The documented input is a folder containing JPG images.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
yassouali/CCT · Official homepage title · retained in its source language
How we review sourcesSuggest a correction