diff --git a/otbtf/examples/tensorflow_v2x/fcnn/fcnn_model.py b/otbtf/examples/tensorflow_v2x/fcnn/fcnn_model.py
index 44285d92bb77695224288cb9d804018c90924c82..4f2377ae9d354b5d6e90728b254de864909eb7fe 100644
--- a/otbtf/examples/tensorflow_v2x/fcnn/fcnn_model.py
+++ b/otbtf/examples/tensorflow_v2x/fcnn/fcnn_model.py
@@ -5,6 +5,7 @@ import logging
import tensorflow as tf
+import otbtf.layers
from otbtf.model import ModelBase
logging.basicConfig(
@@ -23,8 +24,9 @@ INPUT_NAME = "input_xs"
# Name of the output in the `FCNNModel` instance
TARGET_NAME = "predictions"
-# Name (prefix) of the output node in the SavedModel
-OUTPUT_SOFTMAX_NAME = "predictions_softmax_tensor"
+# Name (prefix) of the output nodes in the SavedModel
+OUTPUT_SOFTMAX_NAME = "predictions_softmax"
+OUTPUT_ARGMAX_NAME = "predictions_argmax"
class FCNNModel(ModelBase):
@@ -115,15 +117,22 @@ class FCNNModel(ModelBase):
# from `MyModel.get_output()`. They are two identifiers with a
# different purpose:
# - the output layer name is used only at inference time, to identify
- # the output tensor from which generate the output image,
+ # the output tensor from which generate the output image (e.g.
+ # OUTPUT_SOFTMAX_NAME),
# - the output tensor key identifies the output tensors, mainly to
# fit the targets to model outputs during training process, but it
# can also be used to access the tensors as tf/keras objects, for
# instance to display previews images in TensorBoard.
+ # For convenience, since OTBTF 4.3.0, the post-processed outputs are
+ # automatically added after the output tensor keys. This avoids lazy
+ # users to explicitly name the last layers. When the name is already
+ # took by the layer, the post-processed outputs creating is skipped.
softmax_op = tf.keras.layers.Softmax(name=OUTPUT_SOFTMAX_NAME)
predictions = softmax_op(out_tconv4)
+ argmax_op = otbtf.layers.Argmax()
+ labels = argmax_op(predictions)
- return {TARGET_NAME: predictions}
+ return {TARGET_NAME: predictions, OUTPUT_ARGMAX_NAME: labels}
def dataset_preprocessing_fn(examples: dict):
@@ -146,9 +155,9 @@ def dataset_preprocessing_fn(examples: dict):
"""
return {
INPUT_NAME: examples["input_xs_patches"],
- TARGET_NAME: tf.one_hot(
- tf.squeeze(tf.cast(examples["labels_patches"], tf.int32), axis=-1),
- depth=N_CLASSES
+ TARGET_NAME: otbtf.ops.one_hot(
+ labels=examples["labels_patches"],
+ nb_classes=N_CLASSES
)
}
@@ -173,12 +182,21 @@ def train(params, ds_train, ds_valid, ds_test):
model = FCNNModel(dataset_element_spec=ds_train.element_spec)
# Compile the model
+ # Here using a `dict` to explicitly name the outputs over which the
+ # losses/metrics are computed is a good practice.
model.compile(
- loss=tf.keras.losses.CategoricalCrossentropy(),
+ loss={
+ TARGET_NAME: tf.keras.losses.CategoricalCrossentropy()
+ },
optimizer=tf.keras.optimizers.Adam(
learning_rate=params.learning_rate
),
- metrics=[tf.keras.metrics.Precision(), tf.keras.metrics.Recall()]
+ metrics={
+ TARGET_NAME: [
+ tf.keras.metrics.Precision(class_id=1),
+ tf.keras.metrics.Recall(class_id=1)
+ ]
+ }
)
# Summarize the model (in CLI)
diff --git a/otbtf/model.py b/otbtf/model.py
index 9958510bdf32dd147df273a264714c93d2543ee1..1c648de241bcb4715e17294cd78f8f23dd5b5ab0 100644
--- a/otbtf/model.py
+++ b/otbtf/model.py
@@ -34,7 +34,7 @@ TensorsDict = Dict[str, Tensor]
class ModelBase(abc.ABC):
"""
- Base class for all models
+ Base class for all fully convolutional models
"""
def __init__(
@@ -178,24 +178,35 @@ class ModelBase(abc.ABC):
a dict of post-processed model outputs
"""
-
- # Add extra outputs for inference
+ # Dict of extra outputs for inference
extra_outputs = {}
for out_key, out_tensor in outputs.items():
- for crop in self.inference_cropping:
- extra_output_key = cropped_tensor_name(out_key, crop)
- extra_output_name = cropped_tensor_name(
- out_tensor._keras_history.layer.name, crop
- )
- logging.info(
- "Adding extra output for tensor %s with crop %s (%s)",
- out_key, crop, extra_output_name
+ layer_name = out_tensor._keras_history.layer.name
+ # extra output named after layer
+ srcs_names = [layer_name]
+ if layer_name == out_key:
+ logging.warning(
+ "Output \"%s\" already exist from layer of the same "
+ "name. Skipping extra-outputs creation. If you have "
+ "any doubt, you can use the following command to see "
+ "available model outputs: "
+ "`saved_model_cli show --dir your_model_dir --all`",
+ layer_name
)
- cropped = out_tensor[:, crop:-crop, crop:-crop, :]
- identity = tf.keras.layers.Activation(
- 'linear', name=extra_output_name
- )
- extra_outputs[extra_output_key] = identity(cropped)
+ else:
+ # extra output named after output key
+ srcs_names += [out_key]
+ # Now for all accepted src_name, we create extra outputs
+ for crop in self.inference_cropping:
+ for src_name in srcs_names:
+ tgt_name = cropped_tensor_name(src_name, crop)
+ logging.info(
+ "Adding extra output for tensor %s with crop %s: %s",
+ out_key, crop, tgt_name
+ )
+ eye = tf.keras.layers.Activation("linear", name=tgt_name)
+ extra_out = eye(out_tensor[:, crop:-crop, crop:-crop, :])
+ extra_outputs[f"postproc_{tgt_name}"] = extra_out
return extra_outputs
diff --git a/test/api_unittest.py b/test/api_unittest.py
index 2fe3fe38632d8739dca7a61458ec95d398b0170e..aa2f51b3a146863aab79b74ace1867db0fac7296 100644
--- a/test/api_unittest.py
+++ b/test/api_unittest.py
@@ -8,9 +8,10 @@ from otbtf.examples.tensorflow_v2x.fcnn import create_tfrecords
from otbtf.examples.tensorflow_v2x.fcnn import train_from_patchesimages
from otbtf.examples.tensorflow_v2x.fcnn import train_from_tfrecords
from otbtf.examples.tensorflow_v2x.fcnn.fcnn_model import INPUT_NAME, \
- OUTPUT_SOFTMAX_NAME
+ OUTPUT_SOFTMAX_NAME, OUTPUT_ARGMAX_NAME, TARGET_NAME
from otbtf.model import cropped_tensor_name
-from test_utils import resolve_paths, files_exist, run_command_and_compare
+from test_utils import resolve_paths, files_exist, run_command_and_compare, \
+ run_command_and_test_exist
INFERENCE_MAE_TOL = 10.0 # Dummy value: we don't really care of the mae value but rather the image size etc
@@ -39,42 +40,85 @@ class APITest(unittest.TestCase):
'$TMPDIR/model_from_pimg/variables/variables.index'
]))
- @pytest.mark.order(2)
+ @pytest.mark.order(2, 5)
def test_model_inference1(self):
- self.assertTrue(
- run_command_and_compare(
- command=
+ def _make_command(out_name, rfield, efield):
+ return (
"otbcli_TensorflowModelServe "
"-source1.il $DATADIR/fake_spot6.jp2 "
- "-source1.rfieldx 64 "
- "-source1.rfieldy 64 "
+ f"-source1.rfieldx {rfield} "
+ f"-source1.rfieldy {rfield} "
f"-source1.placeholder {INPUT_NAME} "
"-model.dir $TMPDIR/model_from_pimg "
"-model.fullyconv on "
- f"-output.names {cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 16)} "
- "-output.efieldx 32 "
- "-output.efieldy 32 "
- "-out \"$TMPDIR/classif_model4_softmax.tif?&gdal:co:compress=deflate\" uint8",
- to_compare_dict={
- "$DATADIR/classif_model4_softmax.tif": "$TMPDIR/classif_model4_softmax.tif"},
- tol=INFERENCE_MAE_TOL))
- self.assertTrue(
- run_command_and_compare(
- command=
- "otbcli_TensorflowModelServe "
- "-source1.il $DATADIR/fake_spot6.jp2 "
- "-source1.rfieldx 128 "
- "-source1.rfieldy 128 "
- f"-source1.placeholder {INPUT_NAME} "
- "-model.dir $TMPDIR/model_from_pimg "
- "-model.fullyconv on "
- f"-output.names {cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 32)} "
- "-output.efieldx 64 "
- "-output.efieldy 64 "
- "-out \"$TMPDIR/classif_model4_softmax.tif?&gdal:co:compress=deflate\" uint8",
- to_compare_dict={
- "$DATADIR/classif_model4_softmax.tif": "$TMPDIR/classif_model4_softmax.tif"},
- tol=INFERENCE_MAE_TOL))
+ f"-output.names {out_name} "
+ f"-output.efieldx {efield} "
+ f"-output.efieldy {efield} "
+ "-out \"$TMPDIR/classif_model4_softmax.tif?&gdal:co:compress=deflate\" uint8"
+ )
+
+ command_crop16_softmax = _make_command(
+ out_name=cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 16),
+ rfield=64,
+ efield=32
+ )
+ command_crop32_softmax = _make_command(
+ out_name=cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 32),
+ rfield=128,
+ efield=64
+ )
+ command_crop16_softmax2 = _make_command(
+ out_name=cropped_tensor_name(TARGET_NAME, 16),
+ rfield=64,
+ efield=32
+ )
+ command_crop32_softmax2 = _make_command(
+ out_name=cropped_tensor_name(TARGET_NAME, 32),
+ rfield=128,
+ efield=64
+ )
+ command_crop16_argmax = _make_command(
+ out_name=cropped_tensor_name(OUTPUT_ARGMAX_NAME, 16),
+ rfield=64,
+ efield=32
+ )
+ command_crop32_argmax = _make_command(
+ out_name=cropped_tensor_name(OUTPUT_ARGMAX_NAME, 32),
+ rfield=128,
+ efield=64
+ )
+
+ def _test_compare(command):
+ self.assertTrue(
+ run_command_and_compare(
+ command=command,
+ to_compare_dict={
+ "$DATADIR/classif_model4_softmax.tif":
+ "$TMPDIR/classif_model4_softmax.tif"
+ },
+ tol=INFERENCE_MAE_TOL
+ )
+ )
+
+ def _test_exist(command):
+ self.assertTrue(
+ run_command_and_test_exist(
+ command=command,
+ file_list=["$TMPDIR/classif_model4_softmax.tif"]
+ )
+ )
+
+ # softmax (from layer name)
+ _test_compare(command_crop16_softmax)
+ _test_compare(command_crop32_softmax)
+
+ # softmax (from target key i.e. model output name)
+ _test_compare(command_crop16_softmax2)
+ _test_compare(command_crop32_softmax2)
+
+ # argmax
+ _test_exist(command_crop16_argmax)
+ _test_exist(command_crop32_argmax)
@pytest.mark.order(3)
def test_create_tfrecords(self):
@@ -116,48 +160,6 @@ class APITest(unittest.TestCase):
'$TMPDIR/model_from_tfrecs/variables/variables.index'
]))
- @pytest.mark.order(5)
- def test_model_inference2(self):
- self.assertTrue(
- run_command_and_compare(
- command=
- "otbcli_TensorflowModelServe "
- "-source1.il $DATADIR/fake_spot6.jp2 "
- "-source1.rfieldx 64 "
- "-source1.rfieldy 64 "
- f"-source1.placeholder {INPUT_NAME} "
- "-model.dir $TMPDIR/model_from_pimg "
- "-model.fullyconv on "
- f"-output.names {cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 16)} "
- "-output.efieldx 32 "
- "-output.efieldy 32 "
- "-out \"$TMPDIR/classif_model4_softmax.tif?&gdal:co:compress=deflate\" uint8",
- to_compare_dict={
- "$DATADIR/classif_model4_softmax.tif":
- "$TMPDIR/classif_model4_softmax.tif"
- },
- tol=INFERENCE_MAE_TOL))
-
- self.assertTrue(
- run_command_and_compare(
- command=
- "otbcli_TensorflowModelServe "
- "-source1.il $DATADIR/fake_spot6.jp2 "
- "-source1.rfieldx 128 "
- "-source1.rfieldy 128 "
- f"-source1.placeholder {INPUT_NAME} "
- "-model.dir $TMPDIR/model_from_pimg "
- "-model.fullyconv on "
- f"-output.names {cropped_tensor_name(OUTPUT_SOFTMAX_NAME, 32)} "
- "-output.efieldx 64 "
- "-output.efieldy 64 "
- "-out \"$TMPDIR/classif_model4_softmax.tif?&gdal:co:compress=deflate\" uint8",
- to_compare_dict={
- "$DATADIR/classif_model4_softmax.tif":
- "$TMPDIR/classif_model4_softmax.tif"
- },
- tol=INFERENCE_MAE_TOL))
-
if __name__ == '__main__':
unittest.main()