Gilad Deutsch

Is this implementation useable for tasks where not all of the input is selected, but some subset of the input? I have seen partial implementation (such as the finish_symbol variable),...

my input is torch.Size([1, 3, 160, 160]). Why is the output dimensions torch.Size([1, 1792, 3, 3]) and not 512? I initialize the model like this - resnet = InceptionResnetV1('vggface2').eval()

Why do you define W_1 as a convolutional layer?

2

https://github.com/shirgur/PointerNet/blob/c7ed05066fb40d4f1832dd24de28da259293983c/PointerNet.py#L98 In this line you define W_1 as a convolutional layer whereas in the paper it seems like they mean it to be a Linear layer. Why do you define...

Qurious as to why you use $\phi(y_{t-1}) - \phi(x_{t-1})$ and not use $\nabla_{x_{t-1}}||\phi(y_{t-1}) - \phi(x_{t-1})||_2^2$ similarly to classifier guidance?

Did you consider adding support for running metrics in batches vs. single videos at a time?