Hi, can you provide the code for only implementing transformer of video caption?

Hello,

I am very interested in how you implemented your paper on Spatio-Temporal Graph for Video Captioning with Knowledge Distillation. I hope that you can release your code, soon.

Thanks!

I tried to reproduce your paper, but it's not clear how to adopt the Two-View Transformer in your framework. For the pretrained Faster R-CNN (ResNext-101), did you use the original Visual Genome dataset? Or did you use the filtered version which has less number of categories?

Hi, team! Thanks for your excellent work, although it is not open source.

I have a question about the “Convolutions on the Spatio-Temporal Graph” stage:
The message passing through $G^{time}$ only makes sense when it connects $G^{space}$ through $G^{time}$ and then back to $G^{space}$. Message passing solely within $G^{time}$, or between different $G^{time}$ matrices, doesn’t seem to have any meaningful purpose.

Can someone answer this question? Thanks for your help!

when the code could be released?

Anyone know how to efficiently construct the sparse spatio-temporal graph ?

there are many researchers who want to learn and have an insight from your code.
we all interested in your awesome work.
thus maybe it's highlight time to release your code.
hope it coming soon, thanks~