What model are you using?

[ShramanPramanick]

Hi, I was reviewing your code and found out that your current code uses the CHAN model (AAAI 2020 paper) {https://github.com/jnzs1836/intent-vizor/blob/7f7b72e7f4e9e403ab3ef465b3b32dd09016d9b0/solvers/qfvsum_solver.py#L77}. Will you please confirm if the repo is the most current format?

[jnzs1836]

The used model is here.

[ShramanPramanick]

Thanks for this. However, your current version calls the CHAN model. I also found many errors in relative paths in your script. Will you please make the code readily executable?

Also, I request clarity in these steps - https://github.com/jnzs1836/intent-vizor/blob/7f7b72e7f4e9e403ab3ef465b3b32dd09016d9b0/models/intent_vizor/model.py#L230-L232 Here you are selecting the actual shots using the generated mask. The 'masked_select ' function flattens the batched features into one dimension. Now, you lost information about the number of true shots in each sample. So how do you execute frame_features = frame_features.view(batch.size(0), -1, batch.size(3))? This code will only work for batch_size=1. In that case, batch normalization does not work.

[jnzs1836]

Thanks for finding the bugs!

I will update the code in next week.

On Wed, Jul 13, 2022 at 1:12 AM Shraman Pramanick @.***> wrote:

Thanks for this. However, your current version calls the CHAN model. I also found many errors in relative paths in your script. Will you please make the code readily executable?

Also, I request clarity in these steps - https://github.com/jnzs1836/intent-vizor/blob/7f7b72e7f4e9e403ab3ef465b3b32dd09016d9b0/models/intent_vizor/model.py#L230-L232 Here you are selecting the actual shots. The 'masked_select ' flattens the batched features into one dimension. Now, you lost information about the number of true shots in each sample. So how do you execute frame_features = frame_features.view(batch.size(0), -1, batch.size(3))? This code will only work for batch_size=1. In that case, batch normalization does not work.

— Reply to this email directly, view it on GitHub https://github.com/jnzs1836/intent-vizor/issues/2#issuecomment-1182773857, or unsubscribe https://github.com/notifications/unsubscribe-auth/AENEWKZG7PIPRTPMRV2OAALVTZF25ANCNFSM53MQBDTA . You are receiving this because you commented.Message ID: @.***>

-- Guande Wu

[jnzs1836]

Sorry the code has not been updated due to some policy issues

On Wed, Jul 13, 2022 at 1:13 AM Guan-de Wu @.***> wrote:

Thanks for finding the bugs!

I will update the code in next week.

On Wed, Jul 13, 2022 at 1:12 AM Shraman Pramanick < @.***> wrote:

Thanks for this. However, your current version calls the CHAN model. I also found many errors in relative paths in your script. Will you please make the code readily executable?

Also, I request clarity in these steps - https://github.com/jnzs1836/intent-vizor/blob/7f7b72e7f4e9e403ab3ef465b3b32dd09016d9b0/models/intent_vizor/model.py#L230-L232 Here you are selecting the actual shots. The 'masked_select ' flattens the batched features into one dimension. Now, you lost information about the number of true shots in each sample. So how do you execute frame_features = frame_features.view(batch.size(0), -1, batch.size(3))? This code will only work for batch_size=1. In that case, batch normalization does not work.

— Reply to this email directly, view it on GitHub https://github.com/jnzs1836/intent-vizor/issues/2#issuecomment-1182773857, or unsubscribe https://github.com/notifications/unsubscribe-auth/AENEWKZG7PIPRTPMRV2OAALVTZF25ANCNFSM53MQBDTA . You are receiving this because you commented.Message ID: @.***>

-- Guande Wu

-- Guande Wu

[jnzs1836]

The importing errors of relative errors and other issues are fixed