ScanThink v1.0.0 - Initial Release: Pre-trained(训练权重) Weights & Annotated Gaze Dataset(眼动数据集)

🚀 Project Overview

This is the initial release of ScanThink, a multi-modal research framework integrating human saliency and
temporal gaze data for image recognition.

📦 What's Inside ScanThink_Assets.zip?

This asset bundle includes all the necessary files to reproduce the experiments:

1. Pre-trained Weights (checkpoints/):

• input_saliency_best.pth: Core SaliencyNet for visual attention prediction.
  - think_v6_gemini_best.pth: High-performance multi-modal fusion model.
  - Various baseline and experimental models (V2-V8).

2. Annotated Dataset (data/):
   • gaze_sessions/: Raw human eye-tracking data collected via Tobii.
   • annotation_progress.db: SQLite database mapping gaze data to image categories.
   • images/: A 10-class subset of the COCO dataset used for training/validation.

🛠️ How to Use

1. Download ScanThink_Assets.zip.
2. Extract the contents into your project root directory.
3. Ensure the folder structure matches:

• /checkpoints/*.pth
  - /data/images/
  - /data/gaze_sessions/