Note to Reviewers: This repository contains the official dataset, evaluation code, and data generation scripts for the paper "PedagogyBench: A Cognitive-Driven Benchmark for Multimodal Instructional Video Understanding". All author information has been anonymized for double-blind review.
.
├── 📂 data/
│ ├── 📂 math/ # math QA pairs (json)
│ ├── 📂 physics/ # physics QA pairs (json)
│ └── 📂 biology/ # biology QA pairs (json)
│
├── 📂 video-Segment/ # stage I: Video preprocessing
│ ├── 📜 cut_video_segments.py # Split long videos into clips
│ ├── 📜 segment_with_prompts.py # Pedagogy-driven segmentation logic
│ └── 📜 batch_build_dataset.py # Batch processing utilities
│
├── 📂 annotations/ # Stage II: Multimodal annotation
│ ├── 📜 visual/visual_anno_all_subjects.py # Vision LLM annotation script
│ └── 📜 audio/transcribe_gpu_sentences.py # Audio transcription (ASR) script
│
├── 📂 scripts/ # Helper scripts
│ ├── 📜 fill_metadata_all_subject.py # Merge visual & audio metadata
│ └── 📜 copy_videos_to_label.py # Data organization util
│
├── 📂 QA generation/ # stage III: Question generation
│ ├── 📜 run_pedagogy_gpt4o.py # Main script for generating MCQs/SAQs
│ └── 📂 prompt_templates/ # (Optional) Folder for prompts if applicable
│
├── 📂 evaluation/ # Model evaluation
│ ├── 📂 qwen2.5VL/ # Evaluation code for Qwen2.5-VL
│ ├── 📂 InternVL2.5/ # Evaluation code for InternVL2.5
│ ├── 📂 LLaVa-Next/ # Evaluation code for LlaVA-Nwxt
│ └── ... (other model folders)
│
├── 📜 LICENSE
└── 📜 README.md