Resource exhausted error during training

[DNJ-JINI]

I would like to train the deep learning task semantic segmentation.
Specifications: model with 31 million parameters
image size ( minimum 1024x1024), batch size 256 framework -tensorflow keras 2.7
Total image samples 30,000,
Label Mask samples 30,000
Does colab pro plus allow the training for this specification with out OOM error?

Issue currently facing is even with 2000 number of images and 512x512 size for batch size 32 results in out of memory error in colab pro subscription.