Image Captioning
ResNet-50 + GPT-2 captioner, benchmarked against BLIP and ViT-GPT2
A custom encoder-decoder captioner trained on Flickr30k, behind a unified API that also serves BLIP and ViT-GPT2 for comparison, plus a small web client.
Highlights
- One API for three captioning models
- Custom ResNet-50 encoder feeding a GPT-2 decoder
- Browser client that uploads an image and shows the caption