Quick Start

Prefer a notebook?

See Examples for the same content as runnable Colab notebooks, including a complete pretrained-model finetuning walkthrough.

Building a model

from keras_climate.remote_sensing import UNet, SegFormer
from keras_climate.forecasting import PatchTST
from keras_climate.weather import ConvLSTMNowcaster
from keras_climate.foundation import PrithviClassifier

model = UNet(input_shape=(256, 256, 4), num_classes=5)   # 4-band (e.g. RGB+NIR) input
model.summary()

forecaster = PatchTST(seq_len=336, pred_len=96, num_channels=7)

All builders return a plain keras.Model, so standard model.compile(...) / model.fit(...) / model.save(...) all work unchanged.

Loading real pretrained weights

Several models ship a ready-made loader that downloads a real, publicly hosted checkpoint and returns a model with pretrained weights already loaded — see Pretrained Weights for full coverage:

from keras_climate.weights.pretrained import unet_carvana

model, report = unet_carvana()
model.summary()

Porting your own checkpoint

For models without a built-in loader, use the generic weight-porting framework directly — see Weight Porting for the full guide:

import keras
from keras_climate.remote_sensing import UNet
from keras_climate.weights import WeightConverter, load_torch_state_dict_as_numpy
from keras_climate.weights.mappings import build_unet_mapper

model = UNet(input_shape=(256, 256, 3), num_classes=1)
model(keras.ops.zeros((1, 256, 256, 3)))  # build the model

state_dict = load_torch_state_dict_as_numpy("unet_carvana.pth")
converter = WeightConverter(model, state_dict, build_unet_mapper())
report = converter.convert(strict=False)

model.save_weights("unet_keras.weights.h5")

Or via the CLI wrapper, which bundles a model-build function + name-mapper per checkpoint family in weights/port_weights.py:

python -m keras_climate.weights.port_weights \
    --model unet --checkpoint unet_carvana.pth --output unet_keras.weights.h5