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

Scan

Scan is a Python and PyTorch research implementation of Stacked Cross Attention Network, which compares images and text through cross-attention for image–text matching and retrieval.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Scan — 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 Scan when you need research code for training or evaluating cross-attention image–text matching models, or a basis for retrieval experiments with images and associated captions.

Who it is for

  • Researchers reproducing or extending image–text matching experiments.
  • Developers prototyping image-caption retrieval with image collections and associated text.

Documented uses

  • Train image–text matching models using MS-COCO or Flickr30K data.
  • Evaluate a trained model on a test split.
  • Prepare image-region features for experiments with another dataset.
  • Compare retrieval approaches using different attention directions and aggregation methods.
Before you choose

Requirements & limits

The repository is public and licensed under Apache-2.0.

  • The README recommends Python 2.7, PyTorch 0.3, NumPy greater than 1.12.1, TensorBoard and NLTK’s Punkt tokenizer.
  • Reproducing the described experiments requires dataset files, vocabularies and image features; the README points to a Kaggle download for experiment data.
  • Using a different dataset requires extracting image-region features and converting them to NumPy format.
  • The README documents training with train.py and evaluation with evaluation.evalrank; it does not describe a hosted retrieval interface.

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

Prepare image-region features for a new dataset

Convert image-region features into the NumPy format Scan describes for testing on another dataset.

Before you start

  • Scan repository and its Python/PyTorch dependencies.
  • The bottom-up-attention tools and model for extracting image-region features.
  • Raw images for the dataset.

Steps

  1. Use bottom-up-attention/tools/generate_tsv.py with the bottom-up attention model to extract region features into TSV format.

  2. Use util/convert_data.py to convert the TSV output to a NumPy array.

Expected outcome

Image-region features are converted to the NumPy array format described for use with another dataset.

Guide limitations

  • The README does not provide the full command or configuration for feature extraction in the cited instructions.
  • This covers feature preparation only, not model training or evaluation on the new dataset.
Documentation 1 ↗
Evidence and transparency

Sources & review

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