dialogtekgeek / audiovisualsceneawaredialog Goto Github PK

Hello @dialogtekgeek
I wish to run the model using only the text data. I want to exclude the image features as discussed in your paper. How do I remove the images features from vector z used for training? What changes do I make in the arguments to do so?

Hi dialogtekgeek,

Thanks for your sharing of this code, I am trying to run the code and encounter a problem in Step 2.
`2. Make directory data/charades_features and extract files under the directory from downloaded feature packages:'
It will be highly appreciated that if you can clarify where I can download the feature files or how to extract them.

Thanks.

Hi,

I just run the code provided and I am not able to reproduce the results. I am getting the following:

  name | bleu_1 | bleu_2 | bleu_3 | bleu_4 | meteor | rouger_l | cider
baseline (i3d_rgb,i3d_flow) | 0.247 | 0.161 | 0.112 | 0.081 | 0.096 | 0.227 | 0.648
baseline (i3d_rgb,i3d_flow,vggish) | 0.251 | 0.164 | 0.116 | 0.084 | 0.095 | 0.23 | 0.666

Thoughts?

Thanks!

Thank you for sharing the baseline system.
There is a significant overlap between valid_set4DSTC7-AVSD.json and test_set.json.
Are the reported test_set.json results based on validation on valid_set4DSTC7-AVSD.json?

How could obtain the test.json?