Google translate image

Hey guys, Does any one have an idea about how google’s image translate new feature works ? it inputs text from an image and then outpoots a new image with the translated text, i’m very curious !
Here’s a link

Hi yazid_elbouachi,

One possibilty could be something akin to a transformer for semantic segmentation, where class labels are used to replace existing texts and enter translated texts. The latter requires some additional coding (basically extending text boxes in terms of needed pixels).

But as the saying goes “only Google knows” :slight_smile: