jepa

jepa copied to clipboard

Hi authors, I have a question about this line https://github.com/facebookresearch/jepa/blob/main/evals/video_classification_frozen/utils.py#L74 where we're using the frame-indices to generate the positional encodings. I'm a little confused looking at the clip-indices being over-written...

saunair

Hello, I currently want to use my own dataset using the V-JEPA method to reconstruct images similarly to the way shown in Figure 6b of your paper. Do you have...

alanwang8

Training Loss Increasing After Initial Decrease with Custom Video Dataset

3

Hello Everyone, I've been working with the V-JEPA model for a self-supervised learning project using a custom video dataset. Initially, the training loss decreases as expected, but starts to increase...

DaBihy

Is AutoAugment used for videos or not

1

Hi, According to the paper, autoaugment is only used for the images classification (with the cross attention head), but in the [code](https://github.com/facebookresearch/jepa/blob/main/evals/video_classification_frozen/eval.py#L467) AA is on for the videos as well,...

Separius

PCA feature map visualization of a pre-trained weights look very random, compared to without pre-trained weights loaded

1

Hi, Thank you for this amazing project. I have been exploring the feature maps produced by the pre-trained V-JEPA, using PCA component visualization.  However, the feature maps look very...

icekang