how to get back original state and size

[ramdhan1989]

Hi,
I tried to run cvit_4x4 for navier stoke problem. I would like to know how to get the original state or size of 128 by 128 while the num_querry_point is 1024?

thanks

[sifanexisted]

Apologies for the delayed response. To clarify, the "4x4" refers to the patch size, and the model is trained on images with a resolution of 128x128.

[ramdhan1989]

Hi,
I just want to clarify from eval.py, is the code here will give me the result of 128,128, 3 (height, width, channel)? from my test, the model output is 1024, 3 (num querry point, channel)

l2_error_list = []
for batch in test_loader:
batch = jax.tree_map(lambda x: jnp.array(x), batch)
x, y = batch
pred = model.apply(state.params, x, coords)
pred = pred.reshape(-1, 1, h, w, c)

or

pred = rollout(
state,
x,
coords,
prev_steps=config.dataset.prev_steps,
pred_steps=config.dataset.pred_steps,
rollout_steps=config.eval.rollout_steps,
)

thanks

[sifanexisted]

I think you need to ensure the provided coords is a flattened grid.