Adaptive-Sentinel-Image-Captioning

PyTorch implementation of the Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning by Lu et. al.

Work in Progress.

TO-DO :

• Implement normal LSTM-Attention
• Greedy + Sampling Decoder
• Layer Normalization
• [] Debug Sentinel-LSTM
• [] Implement Beam Search
• [] Tune hyperparameters

Train Loss curve without pre-trained embeddings

Train Loss curve with pretrained embeddings