Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Advancing Assessment Practices in CS Education through AI-Generated Visual Test Cases. In Proceedings of the 2025 Conference on UK and Ireland Computing Education Research (UKICER '25).

License

Notifications You must be signed in to change notification settings

mawady/ukicer25

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Advancing Assessment Practices in CS Education through AI-Generated Visual Test Cases

 Citation

Mohamed Elawady. 2025. Advancing Assessment Practices in CS Education through AI-Generated 
Visual Test Cases. In UK and Ireland Computing Education Research Conference (UKICER 2025), 
September 04–05, 2025, Edinburgh, United Kingdom. ACM, New York, NY, USA, 1 page.
https://doi.org/10.1145/3754508.3754538

Demo

  • Image Classification Validator:

    • Used models: ResNet18, ConvNextV1-Tiny
    • Pre-trained weights: ImageNet
    • Function: Upload an AI-generated image to see top-3 predictions from classification models with probabilities.
    • [Google Colab]
    • [HF Spaces Demo]
  • Image Detection Validator:

    • Used models: YOLOv8n, YOLOv12n
    • Pre-trained weights: COCO
    • Function: Upload an AI-generated image to view the top three predictions from detection models, including their probabilities, along with the corresponding bounding boxes for each model.
    • [Google Colab]
    • [HF Spaces Demo]

Visual Test Cases (sorted from best to worst)

Prompt

  • generate realistic image (cat in room) and write bounding box coordinates of cat within the image to be used later to compare with object detection models. don't show annotation box over the generated image.

OpenAI - ChatGPT GPT5

  • Data Folder (images and detection information in json format file): ./imgs/chatgpt-gpt5/
  • Link: https://chatgpt.com/
  • Discussion: Successfully generated images with bounding box information and text descriptions in JSON format. Further example generation is constrained by paywall restrictions.

Microsoft - M365 Copilot GPT5

  • Data Folder (images and text of LLM reply): ./imgs/copilot-gpt5/
  • Link: https://m365.cloud.microsoft/chat/
  • Discussion: Capable of generating images and text with highlighted detection bounding boxes, but lacks support for exporting results as downloadable files and performs with slow execution speed.

Google - Gemini 2.5 Flash

Black Forest Labs - FLUX.1-dev

StabilityAI - Stable Diffusion 2.1

Craiyon - DALL·E mini

Anthropic - Claude Sonnet 4

  • Data Folder (screenshot of LLM reply): ./imgs/claude-sonnet4/
  • Link: https://claude.ai/chat/
  • Discussion: Unable to automatically generate images or annotation metadata for object detection. However, capable of creating and running an annotation application that allows manual image upload, bounding box drawing, and exporting annotations in multiple formats (COCO, YOLO, Pascal VOC).

About

Advancing Assessment Practices in CS Education through AI-Generated Visual Test Cases. In Proceedings of the 2025 Conference on UK and Ireland Computing Education Research (UKICER '25).

Resources

License

Stars

Watchers

Forks

Contributors 2

  •  
  •