232525 / PureT

Implementation of 'End-to-End Transformer Based Model for Image Captioning' [AAAI 2022]
63 stars 12 forks source link

Want to generate Instagram styled captions #17

Closed shreyassks closed 1 year ago

shreyassks commented 1 year ago

Hi, I just wanted to know if this model architecture works well to generate instagram styled captions??. I am using YFCC100m dataset which has image-caption pairs which serve for my purpose. But when i use this architecture and train it, I don't see the model is able to learn anything. I have used CE loss and am yet to try SCST training. Please suggest

232525 commented 1 year ago

I am sorry that I have no idea about the performance under YFCC100m dataset. But I used to train our model under a Chinese Caption dataset and it get normal performance. And I guess maybe the codes about Dataset and Dataloader are important, so maybe you need to check them to make sure they work correctly.