Picalike Dokuwiki Archive

Image2text

This is an ongoing project to generate text descriptions from a product image.

This system should automatically generate product description texts that are somehow human readable. The aim is to generate a mass of content for online marketing.

Resources

https://pythonawesome.com/image-captioning-using-cnn-and-transformers-in-python/
http://vision.is.tohoku.ac.jp/~kyamagu/papers/yashima2016learning.pdf
https://arxiv.org/abs/2203.15350
https://arxiv.org/abs/2101.10804
http://cs231n.stanford.edu/slides/2021/lecture_11.pdf (p. 82 - pure transformer image captioning)<HTML></ol></HTML>

Currently, we have 'short_tease' as part of crawler metadata. For otto, we have 4,833,786 teasers + images.

References

https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Image-Captioning

Ticket Backup

The idea is, to get images and train a image to text system to automatically generate product description text that are some how human readable. They don't need to be perfect. The aim is to generate a mass of content for online marketing. Example: I choose TVs and get 1000 TV from OTTO. We get the TV image, some metadata and get a text about the product.

Table of Contents

Image2text

Resources

References

Ticket Backup