2026 · AI / ML / Computer Vision
Show and Tell Image Captioning
A PyTorch encoder-decoder that generates captions for images from COCO.
- Input image
- ResNet features and LSTM decoding
- Generated caption
Computer vision projectGitHub ↗
Problem
Generate a short natural-language description of an image.
My role
Implemented training, inference, and evaluation, with CLI and web demo code.
Approach
- Encode images with pretrained ResNet-50 features.
- Train an LSTM decoder with teacher forcing.
- Compare decoding with greedy and beam search and inspect failure cases.
Results
- The repository reports validation BLEU-1 of 0.656 and BLEU-4 of 0.207, along with example captions and failure cases.
Stack
- Python
- PyTorch
- COCO
- BLEU