Add example

MLmodel

@@ -0,0 +1,30 @@
+artifact_path: model
+flavors:
+  python_function:
+    config:
+      device: null
+    data: data
+    env:
+      conda: conda.yaml
+      virtualenv: python_env.yaml
+    loader_module: mlflow.pytorch
+    pickle_module_name: mlflow.pytorch.pickle_module
+    python_version: 3.10.11
+  pytorch:
+    code: null
+    model_data: data
+    pytorch_version: 2.5.0+cu124
+mlflow_version: 2.17.0
+model_size_bytes: 176239334
+model_uuid: 23fc55bd7bea4ce0b7ea15d15fb19d4f
+run_id: 5bc0ffd24c5748ba94f2adc6f33d83f6
+signature:
+  inputs: '[{"type": "tensor", "tensor-spec": {"dtype": "float32", "shape": [1, 3,
+    226, 288]}}]'
+  outputs: '[{"name": "boxes", "type": "tensor", "tensor-spec": {"dtype": "float32",
+    "shape": [71, 4]}}, {"name": "labels", "type": "tensor", "tensor-spec": {"dtype":
+    "int64", "shape": [71]}}, {"name": "scores", "type": "tensor", "tensor-spec":
+    {"dtype": "float32", "shape": [71]}}, {"name": "masks", "type": "tensor", "tensor-spec":
+    {"dtype": "float32", "shape": [71, 1, 226, 288]}}]'
+  params: null
+utc_time_created: '2024-10-23 22:18:33.740363'

conda.yaml

@@ -0,0 +1,16 @@
+channels:
+- conda-forge
+dependencies:
+- python=3.10.11
+- pip<=23.1.2
+- pip:
+  - mlflow==2.17.0
+  - cloudpickle==3.1.0
+  - colorama==0.4.6
+  - defusedxml==0.7.1
+  - nvidia-ml-py==12.560.30
+  - pandas==2.2.3
+  - torch==2.5.0
+  - torchvision==0.20.0
+  - tqdm==4.66.5
+name: mlflow-env

inference.py

@@ -0,0 +1,44 @@
+import torch
+import requests
+import tempfile
+from torchvision.io import read_image
+
+def preprocess(image):
+    from torchvision.io import read_image
+    from torchvision.transforms import v2 as T
+    transforms = []
+
+    transforms.append(T.ToDtype(torch.float, scale=True))
+    transforms.append(T.ToPureTensor())
+    transform =  T.Compose(transforms)
+
+    image = (255.0 * (image - image.min()) / (image.max() - image.min())).to(
+        torch.uint8
+    )
+    image = image[:3, ...]
+    transformed_image = transform(image)
+
+    x = torch.unsqueeze(transformed_image, 0)
+
+    return x # [:3, ...]
+
+def filter_predictions(pred, score_threshold=0.5):
+    keep = pred["scores"] > score_threshold
+    return {k: v[keep] for k, v in pred.items()}
+
+
+model = torch.load("model.pth", map_location=torch.device('cpu'))
+image_url = "https://github.com/cyber2a/Cyber2A-RTS-ToyModel/blob/main/data/images/valtest_nitze_008.jpg?raw=true"
+response = requests.get(image_url)
+with tempfile.NamedTemporaryFile(delete=True, suffix='.jpg') as tmp_file:
+    # Write the content to the temporary file
+    tmp_file.write(response.content)
+    tmp_file_path = tmp_file.name
+    image = read_image(tmp_file_path)
+
+    scaled_tensor = preprocess(image)
+    with torch.no_grad():
+        output = model(scaled_tensor)
+
+
+    print(filter_predictions(output[0]))

python_env.yaml

@@ -0,0 +1,7 @@
+python: 3.10.11
+build_dependencies:
+- pip==23.1.2
+- setuptools==67.7.2
+- wheel==0.40.0
+dependencies:
+- -r requirements.txt

requirements.txt

@@ -0,0 +1,9 @@
+mlflow==2.17.0
+cloudpickle==3.1.0
+colorama==0.4.6
+defusedxml==0.7.1
+nvidia-ml-py==12.560.30
+pandas==2.2.3
+torch==2.2.2
+torchvision==0.17.2
+tqdm==4.66.5