Add sam export scripts for bioengine format (#803)

examples/bioimageio/export_model_for_bioengine.py

@@ -0,0 +1,3 @@
+from micro_sam.bioimageio.bioengine_export import export_bioengine_model
+
+export_bioengine_model("vit_b", "test-export", opset=12)

micro_sam/bioimageio/bioengine_export.py

@@ -0,0 +1,250 @@
+import os
+import warnings
+from typing import Optional, Union
+
+import torch
+from segment_anything.utils.onnx import SamOnnxModel
+
+try:
+    import onnxruntime
+    onnxruntime_exists = True
+except ImportError:
+    onnxruntime_exists = False
+
+from ..util import get_sam_model
+
+
+ENCODER_CONFIG = """name: "%s"
+backend: "pytorch"
+platform: "pytorch_libtorch"
+
+max_batch_size : 1
+input [
+  {
+    name: "input0__0"
+    data_type: TYPE_FP32
+    dims: [3, -1, -1]
+  }
+]
+output [
+  {
+    name: "output0__0"
+    data_type: TYPE_FP32
+    dims: [256, 64, 64]
+  }
+]
+
+parameters: {
+  key: "INFERENCE_MODE"
+  value: {
+    string_value: "true"
+  }
+}"""
+
+
+DECODER_CONFIG = """name: "%s"
+backend: "onnxruntime"
+platform: "onnxruntime_onnx"
+
+parameters: {
+  key: "INFERENCE_MODE"
+  value: {
+    string_value: "true"
+  }
+}
+
+instance_group {
+  count: 1
+  kind: KIND_CPU
+}"""
+
+
+def _to_numpy(tensor):
+    return tensor.cpu().numpy()
+
+
+def export_image_encoder(
+    model_type: str,
+    output_root: Union[str, os.PathLike],
+    export_name: Optional[str] = None,
+    checkpoint_path: Optional[str] = None,
+) -> None:
+    """Export SAM image encoder to torchscript.
+
+    The torchscript image encoder can be used for predicting image embeddings
+    with a backed, e.g. with [the bioengine](https://github.com/bioimage-io/bioengine-model-runner).
+
+    Args:
+        model_type: The SAM model type.
+        output_root: The output root directory where the exported model is saved.
+        export_name: The name of the exported model.
+        checkpoint_path: Optional checkpoint for loading the exported model.
+    """
+    if export_name is None:
+        export_name = model_type
+    name = f"sam-{export_name}-encoder"
+
+    output_folder = os.path.join(output_root, name)
+    weight_output_folder = os.path.join(output_folder, "1")
+    os.makedirs(weight_output_folder, exist_ok=True)
+
+    predictor = get_sam_model(model_type=model_type, checkpoint_path=checkpoint_path)
+    encoder = predictor.model.image_encoder
+
+    encoder.eval()
+    input_ = torch.rand(1, 3, 1024, 1024)
+    traced_model = torch.jit.trace(encoder, input_)
+    weight_path = os.path.join(weight_output_folder, "model.pt")
+    traced_model.save(weight_path)
+
+    config_output_path = os.path.join(output_folder, "config.pbtxt")
+    with open(config_output_path, "w") as f:
+        f.write(ENCODER_CONFIG % name)
+
+
+def export_onnx_model(
+    model_type,
+    output_root,
+    opset: int,
+    export_name: Optional[str] = None,
+    checkpoint_path: Optional[Union[str, os.PathLike]] = None,
+    return_single_mask: bool = True,
+    gelu_approximate: bool = False,
+    use_stability_score: bool = False,
+    return_extra_metrics: bool = False,
+) -> None:
+    """Export SAM prompt encoder and mask decoder to onnx.
+
+    The onnx encoder and decoder can be used for interactive segmentation in the browser.
+    This code is adapted from
+    https://github.com/facebookresearch/segment-anything/blob/main/scripts/export_onnx_model.py
+
+    Args:
+        model_type: The SAM model type.
+        output_root: The output root directory where the exported model is saved.
+        opset: The ONNX opset version.
+        export_name: The name of the exported model.
+        checkpoint_path: Optional checkpoint for loading the SAM model.
+        return_single_mask: Whether the mask decoder returns a single or multiple masks.
+        gelu_approximate: Whether to use a GeLU approximation, in case the ONNX backend
+            does not have an efficient GeLU implementation.
+        use_stability_score: Whether to use the stability score instead of the predicted score.
+        return_extra_metrics: Whether to return a larger set of metrics.
+    """
+    if export_name is None:
+        export_name = model_type
+    name = f"sam-{export_name}-decoder"
+
+    output_folder = os.path.join(output_root, name)
+    weight_output_folder = os.path.join(output_folder, "1")
+    os.makedirs(weight_output_folder, exist_ok=True)
+
+    _, sam = get_sam_model(model_type=model_type, checkpoint_path=checkpoint_path, return_sam=True)
+    weight_path = os.path.join(weight_output_folder, "model.onnx")
+
+    onnx_model = SamOnnxModel(
+        model=sam,
+        return_single_mask=return_single_mask,
+        use_stability_score=use_stability_score,
+        return_extra_metrics=return_extra_metrics,
+    )
+
+    if gelu_approximate:
+        for n, m in onnx_model.named_modules:
+            if isinstance(m, torch.nn.GELU):
+                m.approximate = "tanh"
+
+    dynamic_axes = {
+        "point_coords": {1: "num_points"},
+        "point_labels": {1: "num_points"},
+    }
+
+    embed_dim = sam.prompt_encoder.embed_dim
+    embed_size = sam.prompt_encoder.image_embedding_size
+
+    mask_input_size = [4 * x for x in embed_size]
+    dummy_inputs = {
+        "image_embeddings": torch.randn(1, embed_dim, *embed_size, dtype=torch.float),
+        "point_coords": torch.randint(low=0, high=1024, size=(1, 5, 2), dtype=torch.float),
+        "point_labels": torch.randint(low=0, high=4, size=(1, 5), dtype=torch.float),
+        "mask_input": torch.randn(1, 1, *mask_input_size, dtype=torch.float),
+        "has_mask_input": torch.tensor([1], dtype=torch.float),
+        "orig_im_size": torch.tensor([1500, 2250], dtype=torch.float),
+    }
+
+    _ = onnx_model(**dummy_inputs)
+
+    output_names = ["masks", "iou_predictions", "low_res_masks"]
+
+    with warnings.catch_warnings():
+        warnings.filterwarnings("ignore", category=torch.jit.TracerWarning)
+        warnings.filterwarnings("ignore", category=UserWarning)
+        with open(weight_path, "wb") as f:
+            print(f"Exporting onnx model to {weight_path}...")
+            torch.onnx.export(
+                onnx_model,
+                tuple(dummy_inputs.values()),
+                f,
+                export_params=True,
+                verbose=False,
+                opset_version=opset,
+                do_constant_folding=True,
+                input_names=list(dummy_inputs.keys()),
+                output_names=output_names,
+                dynamic_axes=dynamic_axes,
+            )
+
+    if onnxruntime_exists:
+        ort_inputs = {k: _to_numpy(v) for k, v in dummy_inputs.items()}
+        # set cpu provider default
+        providers = ["CPUExecutionProvider"]
+        ort_session = onnxruntime.InferenceSession(weight_path, providers=providers)
+        _ = ort_session.run(None, ort_inputs)
+        print("Model has successfully been run with ONNXRuntime.")
+
+    config_output_path = os.path.join(output_folder, "config.pbtxt")
+    with open(config_output_path, "w") as f:
+        f.write(DECODER_CONFIG % name)
+
+
+def export_bioengine_model(
+    model_type,
+    output_root,
+    opset: int,
+    export_name: Optional[str] = None,
+    checkpoint_path: Optional[Union[str, os.PathLike]] = None,
+    return_single_mask: bool = True,
+    gelu_approximate: bool = False,
+    use_stability_score: bool = False,
+    return_extra_metrics: bool = False,
+) -> None:
+    """Export SAM model to a format compatible with the BioEngine.
+
+    [The bioengine](https://github.com/bioimage-io/bioengine-model-runner) enables running the
+    image encoder on an online backend, so that SAM can be used in an online tool, or to predict
+    the image embeddings via the online backend rather than on CPU.
+
+    Args:
+        model_type: The SAM model type.
+        output_root: The output root directory where the exported model is saved.
+        opset: The ONNX opset version.
+        export_name: The name of the exported model.
+        checkpoint_path: Optional checkpoint for loading the SAM model.
+        return_single_mask: Whether the mask decoder returns a single or multiple masks.
+        gelu_approximate: Whether to use a GeLU approximation, in case the ONNX backend
+            does not have an efficient GeLU implementation.
+        use_stability_score: Whether to use the stability score instead of the predicted score.
+        return_extra_metrics: Whether to return a larger set of metrics.
+    """
+    export_image_encoder(model_type, output_root, export_name, checkpoint_path)
+    export_onnx_model(
+        model_type=model_type,
+        output_root=output_root,
+        opset=opset,
+        export_name=export_name,
+        checkpoint_path=checkpoint_path,
+        return_single_mask=return_single_mask,
+        gelu_approximate=gelu_approximate,
+        use_stability_score=use_stability_score,
+        return_extra_metrics=return_extra_metrics,
+    )