You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@beam.apache.org by GitBox <gi...@apache.org> on 2022/06/09 20:48:28 UTC
[GitHub] [beam] AnandInguva commented on a diff in pull request #21766: Add Pytorch image segmentation example
AnandInguva commented on code in PR #21766:
URL: https://github.com/apache/beam/pull/21766#discussion_r893892990
##########
sdks/python/apache_beam/examples/inference/pytorch_image_segmentation.py:
##########
@@ -0,0 +1,249 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+""""A pipeline that uses RunInference API to perform image segmentation."""
+
+import argparse
+import io
+import os
+from typing import Iterable
+from typing import Optional
+from typing import Tuple
+
+import apache_beam as beam
+import torch
+from apache_beam.io.filesystems import FileSystems
+from apache_beam.ml.inference.api import PredictionResult
+from apache_beam.ml.inference.api import RunInference
+from apache_beam.ml.inference.pytorch_inference import PytorchModelLoader
+from apache_beam.options.pipeline_options import PipelineOptions
+from apache_beam.options.pipeline_options import SetupOptions
+from PIL import Image
+from torchvision import transforms
+from torchvision.models.detection import maskrcnn_resnet50_fpn
+
+COCO_INSTANCE_CLASSES = [
+ '__background__',
+ 'person',
+ 'bicycle',
+ 'car',
+ 'motorcycle',
+ 'airplane',
+ 'bus',
+ 'train',
+ 'truck',
+ 'boat',
+ 'traffic light',
+ 'fire hydrant',
+ 'N/A',
+ 'stop sign',
+ 'parking meter',
+ 'bench',
+ 'bird',
+ 'cat',
+ 'dog',
+ 'horse',
+ 'sheep',
+ 'cow',
+ 'elephant',
+ 'bear',
+ 'zebra',
+ 'giraffe',
+ 'N/A',
+ 'backpack',
+ 'umbrella',
+ 'N/A',
+ 'N/A',
+ 'handbag',
+ 'tie',
+ 'suitcase',
+ 'frisbee',
+ 'skis',
+ 'snowboard',
+ 'sports ball',
+ 'kite',
+ 'baseball bat',
+ 'baseball glove',
+ 'skateboard',
+ 'surfboard',
+ 'tennis racket',
+ 'bottle',
+ 'N/A',
+ 'wine glass',
+ 'cup',
+ 'fork',
+ 'knife',
+ 'spoon',
+ 'bowl',
+ 'banana',
+ 'apple',
+ 'sandwich',
+ 'orange',
+ 'broccoli',
+ 'carrot',
+ 'hot dog',
+ 'pizza',
+ 'donut',
+ 'cake',
+ 'chair',
+ 'couch',
+ 'potted plant',
+ 'bed',
+ 'N/A',
+ 'dining table',
+ 'N/A',
+ 'N/A',
+ 'toilet',
+ 'N/A',
+ 'tv',
+ 'laptop',
+ 'mouse',
+ 'remote',
+ 'keyboard',
+ 'cell phone',
+ 'microwave',
+ 'oven',
+ 'toaster',
+ 'sink',
+ 'refrigerator',
+ 'N/A',
+ 'book',
+ 'clock',
+ 'vase',
+ 'scissors',
+ 'teddy bear',
+ 'hair drier',
+ 'toothbrush'
+]
+
+COCO_INSTANCE_CLASSES_TO_IDX = {
+ idx: cls
+ for (idx, cls) in enumerate(COCO_INSTANCE_CLASSES)
+}
+
+
+def read_image(image_file_name: str,
+ path_to_dir: Optional[str] = None) -> Tuple[str, Image.Image]:
+ if path_to_dir is not None:
+ image_file_name = os.path.join(path_to_dir, image_file_name)
+ with FileSystems().open(image_file_name, 'r') as file:
+ data = Image.open(io.BytesIO(file.read())).convert('RGB')
+ return image_file_name, data
+
+
+def preprocess_image(data: Image.Image) -> torch.Tensor:
+ image_size = (224, 224)
+ # Pre-trained PyTorch models expect input images normalized with the
+ # below values (see: https://pytorch.org/vision/stable/models.html)
+ normalize = transforms.Normalize(
+ mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+ transform = transforms.Compose([
+ transforms.Resize(image_size),
Review Comment:
Looked at the Segmentation models [doc](https://docs.google.com/spreadsheets/d/10R1nMVOZ3bjLiLRBd6GHS0Sc2ZlY7pShu6tM1KDN2mw/edit#gid=0).
1. The reshape happens internally in the model as well. So we perform two resizes on the images(explicitly here and implicitly in the model code). If we don't resize it, the inputs might be different sizes and during the `batching` in torch `RunInference` API, this would fail.
##########
sdks/python/apache_beam/examples/inference/pytorch_image_segmentation.py:
##########
@@ -0,0 +1,249 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+""""A pipeline that uses RunInference API to perform image segmentation."""
+
+import argparse
+import io
+import os
+from typing import Iterable
+from typing import Optional
+from typing import Tuple
+
+import apache_beam as beam
+import torch
+from apache_beam.io.filesystems import FileSystems
+from apache_beam.ml.inference.api import PredictionResult
+from apache_beam.ml.inference.api import RunInference
+from apache_beam.ml.inference.pytorch_inference import PytorchModelLoader
+from apache_beam.options.pipeline_options import PipelineOptions
+from apache_beam.options.pipeline_options import SetupOptions
+from PIL import Image
+from torchvision import transforms
+from torchvision.models.detection import maskrcnn_resnet50_fpn
+
+COCO_INSTANCE_CLASSES = [
+ '__background__',
+ 'person',
+ 'bicycle',
+ 'car',
+ 'motorcycle',
+ 'airplane',
+ 'bus',
+ 'train',
+ 'truck',
+ 'boat',
+ 'traffic light',
+ 'fire hydrant',
+ 'N/A',
+ 'stop sign',
+ 'parking meter',
+ 'bench',
+ 'bird',
+ 'cat',
+ 'dog',
+ 'horse',
+ 'sheep',
+ 'cow',
+ 'elephant',
+ 'bear',
+ 'zebra',
+ 'giraffe',
+ 'N/A',
+ 'backpack',
+ 'umbrella',
+ 'N/A',
+ 'N/A',
+ 'handbag',
+ 'tie',
+ 'suitcase',
+ 'frisbee',
+ 'skis',
+ 'snowboard',
+ 'sports ball',
+ 'kite',
+ 'baseball bat',
+ 'baseball glove',
+ 'skateboard',
+ 'surfboard',
+ 'tennis racket',
+ 'bottle',
+ 'N/A',
+ 'wine glass',
+ 'cup',
+ 'fork',
+ 'knife',
+ 'spoon',
+ 'bowl',
+ 'banana',
+ 'apple',
+ 'sandwich',
+ 'orange',
+ 'broccoli',
+ 'carrot',
+ 'hot dog',
+ 'pizza',
+ 'donut',
+ 'cake',
+ 'chair',
+ 'couch',
+ 'potted plant',
+ 'bed',
+ 'N/A',
+ 'dining table',
+ 'N/A',
+ 'N/A',
+ 'toilet',
+ 'N/A',
+ 'tv',
+ 'laptop',
+ 'mouse',
+ 'remote',
+ 'keyboard',
+ 'cell phone',
+ 'microwave',
+ 'oven',
+ 'toaster',
+ 'sink',
+ 'refrigerator',
+ 'N/A',
+ 'book',
+ 'clock',
+ 'vase',
+ 'scissors',
+ 'teddy bear',
+ 'hair drier',
+ 'toothbrush'
+]
+
+COCO_INSTANCE_CLASSES_TO_IDX = {
+ idx: cls
+ for (idx, cls) in enumerate(COCO_INSTANCE_CLASSES)
+}
+
+
+def read_image(image_file_name: str,
+ path_to_dir: Optional[str] = None) -> Tuple[str, Image.Image]:
+ if path_to_dir is not None:
+ image_file_name = os.path.join(path_to_dir, image_file_name)
+ with FileSystems().open(image_file_name, 'r') as file:
+ data = Image.open(io.BytesIO(file.read())).convert('RGB')
+ return image_file_name, data
+
+
+def preprocess_image(data: Image.Image) -> torch.Tensor:
+ image_size = (224, 224)
+ # Pre-trained PyTorch models expect input images normalized with the
+ # below values (see: https://pytorch.org/vision/stable/models.html)
+ normalize = transforms.Normalize(
+ mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+ transform = transforms.Compose([
+ transforms.Resize(image_size),
+ transforms.ToTensor(),
+ normalize,
+ ])
+ return transform(data)
+
+
+class PostProcessor(beam.DoFn):
+ def process(self, element: Tuple[str, PredictionResult]) -> Iterable[str]:
+ filename, prediction_result = element
+ prediction_labels = prediction_result.inference['labels']
+ classes = [
+ COCO_INSTANCE_CLASSES_TO_IDX[label.item()]
+ for label in prediction_labels
+ ]
+ yield filename + ';' + str(classes)
+
+
+def parse_known_args(argv):
+ """Parses args for the workflow."""
+ parser = argparse.ArgumentParser()
+ parser.add_argument(
+ '--input',
+ dest='input',
+ default='gs://apache-beam-ml/testing/inputs/'
+ 'it_maskrcnn_resnet50_fpn_coco_validation_inputs.txt',
+ help='Path to the text file containing image names.')
+ parser.add_argument(
+ '--output',
+ dest='output',
+ help='Path where to save output predictions.'
+ ' text file.')
+ parser.add_argument(
+ '--model_state_dict_path',
+ dest='model_state_dict_path',
+ default='gs://apache-beam-ml/'
+ 'models/torchvision.models.detection.maskrcnn_resnet50_fpn.pth',
Review Comment:
Yep. I will make that change
##########
sdks/python/apache_beam/examples/inference/pytorch_image_segmentation.py:
##########
@@ -0,0 +1,249 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+""""A pipeline that uses RunInference API to perform image segmentation."""
Review Comment:
4 double quotes. Would you be able to correct this as well in the `pytorch_image_classification.py` example? or may be I can create a different PR for that change. Anything works
##########
sdks/python/apache_beam/ml/inference/pytorch_inference_it_test.py:
##########
@@ -89,6 +90,42 @@ def test_torch_run_inference_imagenet_mobilenetv2(self):
filename, prediction = prediction.split(',')
self.assertEqual(_EXPECTED_OUTPUTS[filename], prediction)
+ @pytest.mark.uses_pytorch
+ @pytest.mark.it_postcommit
+ def test_torch_run_inference_coco_maskrcnn_resnet50_fpn(self):
+ test_pipeline = TestPipeline(is_integration_test=True)
+ # text files containing absolute path to the coco validation data on GCS
+ file_of_image_names = 'gs://apache-beam-ml/testing/inputs/it_maskrcnn_resnet50_fpn_coco_validation_inputs.txt' # disable: line-too-long
Review Comment:
Hm..I think the model name is not required in the input image name. I don't know why I included the model name as part of the `imagenet data` file.
Can we remove the model name from the input name and have something like `it_coco_validation_inputs.txt`? I will make the change to reflect the change on the example I created. Sorry for the confusion
##########
sdks/python/apache_beam/ml/inference/pytorch_inference_it_test.py:
##########
@@ -89,6 +90,42 @@ def test_torch_run_inference_imagenet_mobilenetv2(self):
filename, prediction = prediction.split(',')
self.assertEqual(_EXPECTED_OUTPUTS[filename], prediction)
+ @pytest.mark.uses_pytorch
+ @pytest.mark.it_postcommit
+ def test_torch_run_inference_coco_maskrcnn_resnet50_fpn(self):
+ test_pipeline = TestPipeline(is_integration_test=True)
+ # text files containing absolute path to the coco validation data on GCS
+ file_of_image_names = 'gs://apache-beam-ml/testing/inputs/it_maskrcnn_resnet50_fpn_coco_validation_inputs.txt' # disable: line-too-long
+ output_file_dir = 'gs://apache-beam-ml/testing/predictions'
+ output_file = '/'.join([output_file_dir, str(uuid.uuid4()), 'result.txt'])
+
+ model_state_dict_path = 'gs://apache-beam-ml/models/torchvision.models.detection.maskrcnn_resnet50_fpn.pth'
+ images_dir = 'gs://apache-beam-ml/datasets/coco/raw-data/val2017'
+ extra_opts = {
+ 'input': file_of_image_names,
+ 'output': output_file,
+ 'model_state_dict_path': model_state_dict_path,
+ 'images_dir': images_dir,
+ }
+ pytorch_image_segmentation.run(
+ test_pipeline.get_full_options_as_args(**extra_opts),
+ save_main_session=False)
+
+ self.assertEqual(FileSystems().exists(output_file), True)
+ predictions = process_outputs(filepath=output_file)
+ actuals_file = 'gs://apache-beam-ml/testing/expected_outputs/test_torch_run_inference_coco_maskrcnn_resnet50_fpn_actuals.txt'
+ actuals = process_outputs(filepath=actuals_file)
+
+ predictions_dict = {}
+ for prediction in predictions:
+ filename, prediction_labels = prediction.split(';')
+ predictions_dict[filename] = prediction_labels
+
+ for actual in actuals:
+ filename, actual_labels = actual.split(';')
+ prediction_labels = predictions_dict[filename]
+ self.assertEqual(actual_labels, prediction_labels)
+
Review Comment:
yes, It's better to put it in a different file like you did. I will work on the change. Thanks
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@beam.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org