neuraltalk2
neuraltalk2 copied to clipboard
Could neuraltalk2 caption images by using other language such as Chinese?
As we know that the grammar is different between English and Chinese, could neuraltalk2 work well if I describe my JPG files by using Chinese?
If you have the training data on Chinese. You can try.
Generate coco_raw.json with captioning in chinese language. Then train the model based on that. After that you will be able to evaluate your images with captioning in chinese language
Anyone tried already for other languages? I need to know if the grammar alignment part needs to be worked or not..
It will caption as per the Chinese sentences you have used to train the model. So if the training model contains correct grammatical sentences then it will caption correct sentences. Please note that I have not tried with the other languages, its my assumption.
Is there a successful example?Can you share examples?
I just generated the coco_raw.json and then I modified to my Language. The prepro script cannot import because it cannot read UTF-8 characters. So the solution needs code change and not straight forward. I did change the prepro script to accept UTF-8 encoding. Yet i have been hit with another road block in training. Will keep posted if something works out further.
Please teach me how to generate coco_raw.json. I have recently explored this side