Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

flow_imagenet checkpoint #102

Open
nik123 opened this issue May 16, 2020 · 0 comments
Open

flow_imagenet checkpoint #102

nik123 opened this issue May 16, 2020 · 0 comments

Comments

@nik123
Copy link

nik123 commented May 16, 2020

The readme states:

The default model has been pre-trained on ImageNet and then Kinetics

As far as I understand, "pre-trained on ImageNet" means 3D convolutional NN which weights are recieved by bootsrapping values from 2D convolutional NN trained on ImageNet. So data/checkpoints/rgb_imagenet checkpoint is RGB network which is initialized with bootstrapped weights from 2D NN (which was trained on RGB images from ImageNet) and then trained on Kinetics dataset. Please, correct me if there are any mistakes.

If the description above is correct then I'm not sure I understand how data/checkpoints/flow_imagenet weights were achieved. I think it is Flow network which is initialized with bootstrapped weights from 2D NN (which was trained on RGB images from ImageNet) and then trained on optical flow values from Kinetics dataset. It's suprising that flow NN is initialized with RGB values. Is it really so?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant