How can we inference on our own images using this repo?

[Mas-Y]

trafficstars

Can anyone suggest how this repo can be used to test on our own images?

[kdexd]

Hi @Mas-Y — this repository provides a baseline model for future comparison. It is a little non-trivial to run inference on arbitrary images, but it's possible — you need to extract image features using https://github.com/nocaps-org/image-feature-extractors before generating captions using this repo. Putting it end-to-end is hard because that repo uses old Caffe, and is encapsulated separately in a docker file.