Tensorflow

research-models.md

Trouble Shooting

Common

• pip 22.x issue

• Compatibility

  • tensorflow-2.18.1
  • tf_keras-2.18.0
  • CUDA® Toolkit 12.3
  • cuDNN SDK 8.9.7

• export PYTHONPATH=$PYTHONPATH:/path/to/tensorflow-models

• tf_keras -> tf.keras

• exp_config.runtime.num_gpus = 1

• To avoid GCS warnings/errors

  • tensorflow_datasets.core.utils.gcs_utils._is_gcs_disabled = True
  • os.environ['NO_GCE_CHECK'] = 'true'

• object_detection/builders/model_builder_tf2_test.py + protobuf + runtime_version issue

  • Step 1: pip install --upgrade protobuf
  • Step 2: save <venv>/lib/python3.10/site-packages/google/protobuf/runtime_version.py somewhere else
  • Step 3: pip install protobuf==3.20.3
  • Step 4: restore runtime_version.py

Windows Native

• Environment

  • OS: Wins 11 x64
    • Shell: PowerShell (PSVersion: 5.1.26100.2161)
  • Python 3.10.9 (venv)
    • Pip 25.0.1

• tensorflow/models

• Command: pip install .

  • setup.py: add 'PyYAML==5.3.1' to REQUIRED_PACKAGES

WSL2

pip install setuptools-scm wheel

• Ref: https://gist.github.com/Ayke/5f37ebdb84c758f57d7a3c8b847648bb

tensorflow_hub.md

Troubleshooting

• Ref: tensorflow/tensorflow#63085 (comment)

mobilenet_v2 ="https://tfhub.dev/google/tf2-preview/mobilenet_v2/classification/4"

classifier_model = mobilenet_v2 #@param ["mobilenet_v2", "inception_v3"] {type:"raw"}
IMAGE_SHAPE = (224, 224)

hub_layer = hub.KerasLayer(classifier_model, input_shape=IMAGE_SHAPE+(3,))
hub_layer_wrapper = tf.keras.layers.Lambda(lambda x: hub_layer(x))
classifier = tf.keras.Sequential([
    hub_layer_wrapper
])