Trained a model by merging RNN and CNN to generate the activity occuring on the image. The model takes pre-trained output from another model, Xception, as image's input and takes another input from text and merges them and generates predictions.
Below is a chart that I prepared for the BLEU score of my VGG16 and Xception model.
The higher the BLEU score, the higher the accuracy of Model.
You can see the Xception Model-4 has the highest score, and that is the model used in this project.
Below are some screenshots, half of them are from training set while others are randomly picked from internet.
As observed below some results are exceptionally good, while some are too bad.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
https://machinelearningmastery.com/
https://data-flair.training/
https://www.youtube.com/