search

happyharrycn / actionformer_release

Code release for ActionFormer (ECCV 2022)
MIT License
434 stars 77 forks source link

some question about mask #21

Closed yilonghe closed 2 years ago

yilonghe commented 2 years ago

Hi! thanks for your great work. I wonder what the out_mask and qx_mask are used for?

tzzcl commented 2 years ago

In general, the mask variable in our code are used to mask the invalid part in our input.

For example, we have a fixed length of 2304 input feature sequences to the network, some feature sequences are shorter than 2304, we will pad them to fixed lengths with 2304, and using the mask to show that some parts are valid, some parts are just the padding part, the mask variable will be passed through the whole network to get the correct mask for each layer.

Specially, out_mask represents the out mask for the output of each layer, and the qx_mask represents the mask for Q in self_attention.

happyharrycn commented 2 years ago

@yilonghe Is this issue addressed by Chenlin's reply?

happyharrycn commented 2 years ago

Closed due to inactivity.