{"id":14959005,"url":"https://github.com/gmalivenko/onnx2keras","last_synced_at":"2025-04-05T23:12:40.525Z","repository":{"id":39652513,"uuid":"193117233","full_name":"gmalivenko/onnx2keras","owner":"gmalivenko","description":"Convert ONNX model graph to Keras model format.","archived":false,"fork":false,"pushed_at":"2024-06-20T09:42:46.000Z","size":179,"stargazers_count":201,"open_issues_count":111,"forks_count":116,"subscribers_count":6,"default_branch":"master","last_synced_at":"2025-03-29T22:07:49.885Z","etag":null,"topics":["deep-convolutional-networks","deep-learning","keras","onnx","onnx2keras","tensorflow","tensorflow-models"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/gmalivenko.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2019-06-21T15:02:28.000Z","updated_at":"2025-03-06T08:36:54.000Z","dependencies_parsed_at":"2024-06-18T22:57:21.571Z","dependency_job_id":"87e276be-ed63-48db-8dea-f344fde97883","html_url":"https://github.com/gmalivenko/onnx2keras","commit_stats":null,"previous_names":["nerox8664/onnx2keras"],"tags_count":23,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/gmalivenko%2Fonnx2keras","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/gmalivenko%2Fonnx2keras/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/gmalivenko%2Fonnx2keras/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/gmalivenko%2Fonnx2keras/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/gmalivenko","download_url":"https://codeload.github.com/gmalivenko/onnx2keras/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247411239,"owners_count":20934653,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["deep-convolutional-networks","deep-learning","keras","onnx","onnx2keras","tensorflow","tensorflow-models"],"created_at":"2024-09-24T13:18:40.905Z","updated_at":"2025-04-05T23:12:40.503Z","avatar_url":"https://github.com/gmalivenko.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# onnx2keras\n\nONNX to Keras deep neural network converter. \n\n[![GitHub License](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)\n[![Python Version](https://img.shields.io/badge/python-2.7%2C3.6-lightgrey.svg)](https://github.com/gmalivenko/onnx2keras)\n[![Downloads](https://pepy.tech/badge/onnx2keras)](https://pepy.tech/project/onnx2keras)\n![PyPI](https://img.shields.io/pypi/v/onnx2keras.svg)\n\n## Requirements\n\nTensorFlow 2.0\n\n## API\n\n`onnx_to_keras(onnx_model, input_names, input_shapes=None, name_policy=None, verbose=True, change_ordering=False) -\u003e {Keras model}`\n\n`onnx_model`: ONNX model to convert\n\n`input_names`: list with graph input names\n\n`input_shapes`: override input shapes (experimental)\n\n`name_policy`: ['renumerate', 'short', 'default'] override layer names (experimental)\n\n`verbose`: detailed output\n\n`change_ordering:` change ordering to HWC (experimental)\n\n\n## Getting started\n\n### ONNX model\n```python\nimport onnx\nfrom onnx2keras import onnx_to_keras\n\n# Load ONNX model\nonnx_model = onnx.load('resnet18.onnx')\n\n# Call the converter (input - is the main model input name, can be different for your model)\nk_model = onnx_to_keras(onnx_model, ['input'])\n```\n\nKeras model will be stored to the `k_model` variable. So simple, isn't it?\n\n\n### PyTorch model\n\nUsing ONNX as intermediate format, you can convert PyTorch model as well.\n\n```python\nimport numpy as np\nimport torch\nfrom torch.autograd import Variable\nfrom pytorch2keras.converter import pytorch_to_keras\nimport torchvision.models as models\n\nif __name__ == '__main__':\n    input_np = np.random.uniform(0, 1, (1, 3, 224, 224))\n    input_var = Variable(torch.FloatTensor(input_np))\n    model = models.resnet18()\n    model.eval()\n    k_model = \\\n        pytorch_to_keras(model, input_var, [(3, 224, 224,)], verbose=True, change_ordering=True)\n\n    for i in range(3):\n        input_np = np.random.uniform(0, 1, (1, 3, 224, 224))\n        input_var = Variable(torch.FloatTensor(input_np))\n        output = model(input_var)\n        pytorch_output = output.data.numpy()\n        keras_output = k_model.predict(np.transpose(input_np, [0, 2, 3, 1]))\n        error = np.max(pytorch_output - keras_output)\n        print('error -- ', error)  # Around zero :)\n```\n\n### Deplying model as frozen graph\n\nYou can try using the snippet below to convert your onnx / PyTorch model to frozen graph. It may be useful for deploy for Tensorflow.js / for Tensorflow for Android / for Tensorflow C-API.\n\n```python\nimport numpy as np\nimport torch\nfrom pytorch2keras.converter import pytorch_to_keras\nfrom torch.autograd import Variable\nimport tensorflow as tf\nfrom tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2\n\n\n# Create and load model\nmodel = Model()\nmodel.load_state_dict(torch.load('model-checkpoint.pth'))\nmodel.eval()\n\n# Make dummy variables (and checking if the model works)\ninput_np = np.random.uniform(0, 1, (1, 3, 224, 224))\ninput_var = Variable(torch.FloatTensor(input_np))\noutput = model(input_var)\n\n# Convert the model!\nk_model = \\\n    pytorch_to_keras(model, input_var, (3, 224, 224), \n                     verbose=True, name_policy='short',\n                     change_ordering=True)\n\n# Save model to SavedModel format\ntf.saved_model.save(k_model, \"./models\")\n\n# Convert Keras model to ConcreteFunction\nfull_model = tf.function(lambda x: k_model(x))\nfull_model = full_model.get_concrete_function(\n    tf.TensorSpec(k_model.inputs[0].shape, k_model.inputs[0].dtype))\n\n# Get frozen ConcreteFunction\nfrozen_func = convert_variables_to_constants_v2(full_model)\nfrozen_func.graph.as_graph_def()\n\nprint(\"-\" * 50)\nprint(\"Frozen model layers: \")\nfor layer in [op.name for op in frozen_func.graph.get_operations()]:\n    print(layer)\n\nprint(\"-\" * 50)\nprint(\"Frozen model inputs: \")\nprint(frozen_func.inputs)\nprint(\"Frozen model outputs: \")\nprint(frozen_func.outputs)\n\n# Save frozen graph from frozen ConcreteFunction to hard drive\ntf.io.write_graph(graph_or_graph_def=frozen_func.graph,\n                  logdir=\"./frozen_models\",\n                  name=\"frozen_graph.pb\",\n                  as_text=False)\n```\n\n\n## License\nThis software is covered by MIT License.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fgmalivenko%2Fonnx2keras","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fgmalivenko%2Fonnx2keras","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fgmalivenko%2Fonnx2keras/lists"}