Image Captioning v1

Image Captioning using VGG-RNN model (with experimental Swin Attention Gated model).

NOTE: If this sparks any interests in me then I will make it performant.

Installation

pip install -r requirements.txt

Train Model

python3 main.py --train --epoch <num>

Evaluate Model

python3 main.py --test

Download dataset

Download dataset inside data directory. All internal files are designed to work with COCO directory being in data directory.

cd data
python3 download_dataset.py

Results

Metric	Score
Bleu_1	0.5953289327600356
Bleu_2	0.40645970649523894
Bleu_3	0.26734383453626503
Bleu_4	0.17662071952569972
METEOR	0.18679035615274345
ROUGE_L	0.4412577702224827
CIDEr	0.5250028088906024