Contact person: ViT-DNE Interface Development & Migration to CiRA CORE

Phone:Show

Email:Show

Location: Srinagar, India

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"This project focuses on developing a continual learning-based industrial anomaly detection system using the MVTec AD dataset. The core objective is to address catastrophic forgetting in sequential learning when detecting new defect types in industrial objects.

The system is designed in two phases:

Python Implementation (Research-Oriented) – Building a model using Vision Transformer (ViT) as the backbone with Distribution Normalization Encoding (DNE) for anomaly scoring. The model is trained sequentially on subsets of industrial defect classes and evaluated using AUC and accuracy matrices to measure retention and generalization across tasks.

CiRA CORE Implementation (Applied/No-Code) – Migrating the Python-based continual learning pipeline into CiRA CORE, a no-code AI platform, to visually replicate dataset handling, model training, evaluation, and prediction, making it accessible without writing code.

This hybrid workflow ensures both a strong research foundation (Python/ViT-DNE) and a practical deployment (CiRA CORE).

Main Tasks

Dataset Handling

Load and preprocess the MVTec AD dataset.

Split into sequential tasks for continual anomaly detection.

Python Model Development

Implement a Vision Transformer (ViT) backbone.

Integrate DNE (Distribution Normalization Encoding) for anomaly scoring.

Build a continual learning loop with evaluation metrics (AUC, accuracy matrix, t-SNE, histograms).

Testing & Evaluation

Run training across sequential tasks.

Log and visualize performance degradation and retention.

CiRA CORE Migration

Recreate the pipeline inside CiRA CORE using visual layers.

Import dataset, configure preprocessing, and rebuild model architecture.

Train and evaluate directly in CiRA CORE.

Provide prediction capability for unseen samples.

Documentation

Deliver final Python source code.

Provide a CiRA CORE project file.

Write a user-friendly guide for running both versions.

Skills Required

Python (PyTorch / TensorFlow / Keras) – for model development.

Knowledge of Transformers (ViT) and anomaly detection methods (DNE).

Familiarity with Continual/Lifelong Learning techniques.

Experience with CiRA CORE (or similar no-code AI tools).

Data preprocessing, visualization (t-SNE, histograms), and evaluation metrics (AUC, Accuracy).

Strong debugging and documentation skills." (client-provided description)