Hi, I’m working on a customized model-based agent based on the Dreamer implementation in Pytorch.
I found out that adding the 5-rank tensor of shape [B, T, 3, H, W] generates the gif in the Tensorboard X visualization, and
the code implementation regarding this seems to be here.
In my thought, If I am to add additional visualization on images (not gifs) having the shape of [3, H, W], I should modify the TBXLogger class .
I am curious if the manual modification of TBXLogger class is okay with the future feature updates of this repo.
If not, what could be the best practice If I want to visualize additional data on the TensorboardX?
(I may add a visualization of the voxel, point clouds, etc.)
Thanks in advance for your help!