Hi. I am currently trying to reproduce the result from the paper. However I notice that the contribution masks are all black from the dense motion network. According to the paper dropout is used to solve this issue. However my experiments show that sometimes some contribution masks can be generated. May I ask what may impact the generation of contribution masks? (e.g. batch size, dropout p, learning rate, luck etc.). Lastly, thank you for your brilliant work.
Hi. I am currently trying to reproduce the result from the paper. However I notice that the contribution masks are all black from the dense motion network. According to the paper dropout is used to solve this issue. However my experiments show that sometimes some contribution masks can be generated. May I ask what may impact the generation of contribution masks? (e.g. batch size, dropout p, learning rate, luck etc.). Lastly, thank you for your brilliant work.