Skip to content

koles289/udacity-Image-Captioning

Repository files navigation

UDACITY (Computer Vision Nanodegree) - Image Captioning

In this project, we build a neural network based on the following paper https://arxiv.org/pdf/1411.4555.pdf.
Architecture of net consists of two parts, Encoder and Decoder. Encoder serves as a feature extractor and decoder serves for generating sequences. In this project, the decoder consisted of LSTM layers with dropout.
The net was trained on captions for MS COCO dataset.
For evaluation of neural network performance, we used perplexity. It measures how well the neural network predits the sample. A low perplexity means that NN is good at predicting captions for samples.


The result of training was automatically generated captios for sample images that were in many cases quite accurate...


but in some cases it was completly wrong.


Obvioulsly, the neural network can not generate accurate caption for objects that it sees for the first time and the result of this behavior are many funny captions...

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published