Closed shiyi-z closed 1 year ago
I'm really admire the work you have done! Especially the work on detecter head, it's a very nice idea. If now i want use the trident-head in other model, such as actionformer, what should i do. Because i find that there are two parameters(boundary_kernel_size and num_bins) not contained in actionformer. How should i process the two parameters in Actionformer? Is it work to simply add the parameter to Actionformer? I would be very grateful if you could help me with this!
Hi, we build our code base on the Actionformer. If you want to use trident-head, you should focus on the part with self.use_trident_head
in meta_archs.py
. We predict the bin set of each instant rather then their offset directly, and then they will be calculated to the offset with decode_offset
function.
I'm really admire the work you have done! Especially the work on detecter head, it's a very nice idea. If now i want use the trident-head in other model, such as actionformer, what should i do. Because i find that there are two parameters(boundary_kernel_size and num_bins) not contained in actionformer. How should i process the two parameters in Actionformer? Is it work to simply add the parameter to Actionformer? I would be very grateful if you could help me with this!