Deep Learning · Academic
Image Captioning System
A deep-learning model that generates natural-language descriptions of images, trained on MS-COCO.
My final-year Big Data & AI deliverable at CESI: an end-to-end image captioning pipeline.
Architecture
- CNN encoder extracts visual features from each image.
- LSTM decoder generates a caption word by word.
- An attention mechanism lets the decoder focus on the relevant region of the image at each step, improving caption quality.
Training
Trained on the MS-COCO dataset, with evaluation on standard captioning metrics (BLEU).
Takeaway
Beyond the model itself, this project sharpened my understanding of the full ML lifecycle — data preparation, training, evaluation — which is increasingly relevant as AI systems fall under regulatory scrutiny (see my work on the EU AI Act).