Pretrained model with R(2+1) D backbone from Table 10

[fmthoker]

Thanks for releasing the code and pretrained models of your amazing work "Audio-Visual Instance Discrimination with Cross-Modal Agreement". I noticed that you used different architectures for R(2+1)D in different experiments as shown in Table 9 and Table 10. Can you please release/share the kinetics pretrained model where you used the architecture of Table 10 for the R(2+1) D backbone?
I am working on self-supervised learning and want to include your paper in my current project. For comparison purposes, I want to use the same backbone as done by previous works.

[pedro-morgado]

Hi, thanks for your interest in our work. The models described in Table 10 have already been released in this repo.
Your confusion seems to come from the fact that there's a typo in the captions of Tables 9 and 10. The architecture of Table 10 is for the large-scale experiments (in Section 5), and that of table 9 is for the rest (Section 3 and 4). Sorry about the confusion.