Yi Yang

Hi all, I tried to reproduce your error according to "I have a long video of length, say, (100, 224, 224, 3). I reshape it to (100, 256, 256, 3)...

We attempt to reproduce your reported issues and here it is. It took approximately 40 minutes to see the first training log result (Also the codebase uses CPU to train...

@TahaRazzaq From the screenshot, I don't see the training stops. Could you verify if the training message just hang there (if hanging there, could you just wait for i.e. 1...

Hi vitacon, Unfortunately we don't have experinece running the TAP models in this repository on Window. However, there is another related issue, please take a look if that is helpful...

`pip install kubric` is not recommended. Instead, try download kubric from github as a subdirectory and place it as a separate directory beside tapnet. ``` git clone https://github.com/google-research/kubric.git Install requirements:...

Note that in the colab notebook, the query points are randomly sampled with each run, which may result in different visualizations. To ensure consistency with the results presented in the...

The Colab notebook aims to provide a minimal demo for illustrative purposes. Loading the TAP-Vid DAVIS dataset with groundtruth tracks or selecting points manually should not be too complicated. To...

@zacharyfmarion We just updated the colab demos with the ability to select points yourself and run the prediction. Please check the last two cells as an example.

Hi Shivam, The differences you observed in the figure and video generation compared to the final release could be attributed to the usage of different checkpoints we used during the...

Hi Shivam, Glad it works. And good observations. Points on the legs are currently much harder to track due to 1. thinness 2. faster motion 3. fewer discriminative cues around....