scarescrow / storyreel

Generating short stories from images

StoryReel

This project attempts to generate stories from images. It achieves this by combining multiple state-of-the-art neural networks for the tasks of face detection, face recognition, object detection and text generation. It also uses NER and retrieval.

Models Used

Task	Technique
Face Detection	MTCNN
Face Recognition	ResNet-50 (trained on VGGFace2)
Object Detection	VGG16 (trained on ImageNet)
Story Retrieval	Cosine Similarity
NER	Spacy
Story Generation	GPT-2 (trained on ROCStories)

Flow Diagram

Report and Poster

The report is uploaded here, and a poster demonstrating the project can be found here.

Sample Outputs

Sample Story 1

Sample Story 2