Image Captioning

Model using Inception Net V3, LSTM and Glove (Using SSD300 to improve feature)

How to run?

Install lib

pip install -r requirements.txt

Download data

python data_download.py

Note : Dataset is MS COCO 2014 and Glove <Wikipedia 2014 + Gigaword 5>. This is large dataset, long download.

Preprocess

You must run

python preprocess.py

With COCO datase this command runs for a long time you can download and coppy them to ROOT / process_data

Download here

Trainning

python train.py --batch-size 64 --output weights --epochs 30

You can download pre-train model an copy them to ROOT / weights

Download here

Predict

python predict.py --image path/to/image --weight path/to/weight --k-beam 9

Result

You can see sumary in summary.ipynb