From e22c0981e632cc2a956f24685d53a9b6788e69c7 Mon Sep 17 00:00:00 2001 From: Starbeamrainbowlabs Date: Fri, 9 Dec 2022 18:35:17 +0000 Subject: [PATCH] actually use dice loss --- aimodel/src/lib/ai/model_rainfallwater_mono.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/aimodel/src/lib/ai/model_rainfallwater_mono.py b/aimodel/src/lib/ai/model_rainfallwater_mono.py index b3ec8cb..a48514e 100644 --- a/aimodel/src/lib/ai/model_rainfallwater_mono.py +++ b/aimodel/src/lib/ai/model_rainfallwater_mono.py @@ -7,6 +7,7 @@ from .components.convnext import make_convnext from .components.convnext_inverse import do_convnext_inverse from .components.LayerStack2Image import LayerStack2Image from .components.LossCrossentropy import LossCrossentropy +from .components.LossDice import LossDice def model_rainfallwater_mono(metadata, model_arch_enc="convnext_xtiny", model_arch_dec="convnext_i_xtiny", feature_dim=512, batch_size=64, water_bins=2, learning_rate=None, heightmap_input=False): """Makes a new rainfall / waterdepth mono model. @@ -87,7 +88,8 @@ def model_rainfallwater_mono(metadata, model_arch_enc="convnext_xtiny", model_ar optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate) model.compile( optimizer=optimizer, - loss=LossCrossentropy(batch_size=batch_size), + # loss=LossCrossentropy(batch_size=batch_size), + loss=LossDice(), # loss=tf.keras.losses.CategoricalCrossentropy(), metrics=[tf.keras.metrics.CategoricalAccuracy()] )