add efficientdets

DariaMityagina · Nov 5, 2020 · 5944695 · 5944695
1 parent 0476ba9
commit 5944695
Show file tree

Hide file tree

Showing 13 changed files with 738 additions and 0 deletions.
diff --git a/models/public/efficientdet-d0-tf/accuracy-check.yml b/models/public/efficientdet-d0-tf/accuracy-check.yml
@@ -0,0 +1,23 @@
+models:
+  - name: efficientdet-d0-tf
+    launchers:
+      - framework: dlsdk
+        adapter: ssd
+    datasets:
+      - name: ms_coco_detection_90_class_without_backgound
+        preprocessing:
+          - type: resize
+            aspect_ratio_scale: fit_to_window
+            size: 512
+          - type: padding
+            size: 512
+            pad_type: right_bottom
+
+        postprocessing:
+          - type: faster_rcnn_postprocessing_resize
+            size: 512
+          - type: shift_labels
+            offset: 1
+
+        metrics:
+          - type: coco_precision
diff --git a/models/public/efficientdet-d0-tf/efficientdet-d0-tf.md b/models/public/efficientdet-d0-tf/efficientdet-d0-tf.md
@@ -0,0 +1,86 @@
+# efficientdet-d0-tf
+
+## Use Case and High-Level Description
+
+The "efficientdet-d0" model is one of the [EfficientDet](https://arxiv.org/abs/1911.09070)
+models  designed to perform object detection. This model was pretrained in TensorFlow*.
+All the EfficientDet models have been pretrained on the MSCOCO* image database.
+For details about this family of models, check out the Google AutoML [repository]
+(https://github.com/google/automl/tree/master/efficientdet).
+
+## Example
+
+## Specification
+
+| Metric            | Value           |
+|-------------------|-----------------|
+| Type              | Object detection|
+| GFLOPs            | 0.819           |
+| MParams           | 5.268           |
+| Source framework  | TensorFlow\*    |
+
+## Accuracy
+
+| Metric | Converted model |
+| ------ | --------------- |
+| coco_precision | 31.95%|
+
+## Performance
+
+## Input
+
+### Original Model
+
+Image, name - `convert_image/Cast`,  shape - `[1x512x512x3]`, format is `[BxHxWxC]`, where:
+
+- `B` - batch size
+- `H` - height
+- `W` - width
+- `C` - channel
+
+Channel order is `RGB`.
+
+### Converted Model
+
+Image, name - `convert_image/Cast/placeholder_port_0`,  shape - `[1x3x512x512]`, format is `[BxCxHxW]`, where:
+
+- `B` - batch size
+- `C` - channel
+- `H` - height
+- `W` - width
+
+Channel order is `BGR`.
+
+## Output
+
+### Original Model
+
+The array of summary detection information, name: `detections`, shape: [1, 7, N], where N is the number of detected
+bounding boxes. For each detection, the description has the format:
+[`image_id`, `y_min`, `x_min`, `y_max`, `x_max`, `confidence`, `label`],
+    where:
+
+    - `image_id` - ID of the image in the batch
+    - (`x_min`, `y_min`) - coordinates of the top left bounding box corner
+    - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner
+    - `confidence` - confidence for the predicted class
+    - `label` - predicted class ID, starting from 1
+
+### Converted Model
+
+The array of summary detection information, name: `detections`, shape: [1, 1, N, 7], where N is the number of detected
+bounding boxes. For each detection, the description has the format:
+[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
+    where:
+
+    - `image_id` - ID of the image in the batch
+    - `label` - predicted class ID, starting from 0
+    - `conf` - confidence for the predicted class
+    - (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
+    - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner  (coordinates stored in normalized format, in range [0, 1])
+
+## Legal Information
+
+The original model is distributed under the
+[Apache License, Version 2.0](https://raw.githubusercontent.com/google/automl/master/LICENSE).
+A copy of the license is provided in [APACHE-2.0-TF-AutoML.txt](../licenses/APACHE-2.0-TF-AutoML.txt).
diff --git a/models/public/efficientdet-d0-tf/model.yml b/models/public/efficientdet-d0-tf/model.yml
@@ -0,0 +1,113 @@
+# Copyright (c) 2020 Intel Corporation
+#
+# Licensed 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.
+
+description: >-
+  The "efficientdet-d0" model is one of the EfficientDet <https://arxiv.org/abs/1911.09070>
+  models  designed to perform object detection. This model was pretrained in TensorFlow*.
+  All the EfficientDet models have been pretrained on the MSCOCO* image database.
+  For details about this family of models, check out the Google AutoML repository
+  <https://github.com/google/automl/tree/master/efficientdet>.
+task_type: detection
+files:
+  - name: efficientdet-d0.tar.gz
+    size: 28828936
+    sha256: 74794c937aa1fa2f559c2393d22b251b1f7135a49b108bd0414bc4f4800ca15d
+    source: https://storage.googleapis.com/cloud-tpu-checkpoints/efficientdet/coco2/efficientdet-d0.tar.gz
+  - name: model_inspect.py
+    size: 20388
+    sha256: 6c68fe02f10d62dd87c2fc550b41c3df5cce52f0449be22e0699fbf209e3cbc1
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/model_inspect.py
+  - name: hparams_config.py
+    size: 13750
+    sha256: a7f9a3215a864e2f393addefc997ceba1d78ccba4909390310c453e391c9710b
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/hparams_config.py
+  - name: inference.py
+    size: 25076
+    sha256: 1f0a633de186f9b786979ead00921b910e9853bb33717328f76c1f71af8be997
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/inference.py
+  - name: dataloader.py
+    size: 19251
+    sha256: 5e4bdfc1a746dd2fffe6a0d7bd83ea6422b2883fce5e8e7fb7ed381e919c11a9
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/dataloader.py
+  - name: nms_np.py
+    size: 11404
+    sha256: e72ffbab850f9a267f9cc5ae480a4ee0936402fd109dfd3c2fa7b5644e71245f
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/nms_np.py
+  - name: utils.py
+    size: 25677
+    sha256: eb68091efcf989a908225c2af6d7f632f90c6602659ab06dea628cbdcabdd403
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/utils.py
+  - name: keras/efficientdet_keras.py
+    size: 32414
+    sha256: 4b60b29aabae4365bb9d77b40698a974de9fd4d998f9be94df12b184c0e09ca9
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/keras/efficientdet_keras.py
+  - name: backbone/efficientnet_builder.py
+    size: 11609
+    sha256: 868b9b4cd06c39e1ec15ea1005c36771676f30003c034a0bed4d06e00932905c
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/backbone/efficientnet_builder.py
+  - name: backbone/efficientnet_model.py
+    size: 27419
+    sha256: 0a488e70c46bfa6d98fec66b4c518e0b3a25eedd8ca836123f4ef335dd12bf0c
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/backbone/efficientnet_model.py
+  - name: backbone/backbone_factory.py
+    size: 2963
+    sha256: 3babe95f1bd104fdb1f54e49e68e2a12478933058d6a09c2a46401e80fab06c9
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/backbone/backbone_factory.py
+  - name: keras/anchors.py
+    size: 9175
+    sha256: a88000a453e1ec1194cff2155a1148ef1c2ca87ad563cff2f412f1a597c3ffd7
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/keras/anchors.py
+  - name: keras/util_keras.py
+    size: 2198
+    sha256: 81e21202054018d7616ab1d0f41eceaf8b4e31a6d6ee88cb53c251abea9fa582
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/keras/util_keras.py
+  - name: keras/fpn_configs.py
+    size: 6120
+    sha256: 245cdd3308141fbb6d2b04b850d45fabb56a0f8ac74c857c7dd0d38238664f0d
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/keras/fpn_configs.py
+  - name: keras/postprocess.py
+    size: 17677
+    sha256: 7ca776128887a813089dfc81a61cd4e82b4c3add100a5bafc1951f7dedd1fb4b
+    source: https://raw.githubusercontent.com/google/automl/341af7d4da7805c3a874877484e133f33c420ec5/efficientdet/keras/postprocess.py
+postprocessing:
+  # disable imports that aren't needed for this model and code that uses them
+  - $type: regex_replace
+    file: inference.py
+    pattern: '(import det_model_fn|from (keras import label_util|visualize))'
+    replacement: '# \g<0>'
+  - $type: regex_replace
+    file: backbone/backbone_factory.py
+    pattern: 'from backbone import efficientnet_lite_builder'
+    replacement: '# \g<0>'
+  - $type: regex_replace
+    file: dataloader.py
+    pattern: 'from (object_detection|keras)'
+    replacement: '# \g<0>'
+  - $type: regex_replace
+    file: keras/anchors.py
+    pattern: 'from object_detection'
+    replacement: '# \g<0>'
+
+  - $type: unpack_archive
+    format: gztar
+    file: efficientdet-d0.tar.gz
+model_optimizer_args:
+  - --input_shape=[1,512,512,3]
+  - --input=convert_image/Cast
+  - --reverse_input_channels
+  - --input_model=$conv_dir/efficientdet-d0_saved_model/efficientdet-d0_frozen.pb
+  - --output=concat,concat_1
+  - --transformations_config=$mo_dir/extensions/front/tf/automl_efficientdet.json
+framework: tf
+license: https://raw.githubusercontent.com/google/automl/master/LICENSE
diff --git a/models/public/efficientdet-d0-tf/pre-convert.py b/models/public/efficientdet-d0-tf/pre-convert.py
@@ -0,0 +1,38 @@
+#!/usr/bin/env python3
+
+# Copyright (c) 2020 Intel Corporation
+#
+# Licensed 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.
+
+import argparse
+import subprocess
+import sys
+
+from pathlib import Path
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('input_dir', type=Path)
+    parser.add_argument('output_dir', type=Path)
+    args = parser.parse_args()
+
+    subprocess.run([sys.executable, '--',
+        str(args.input_dir / 'model_inspect.py'),
+        "--runmode=saved_model",
+        "--model_name=efficientdet-d0",
+        "--ckpt_path={}".format(args.input_dir / "efficientdet-d0"),
+        "--saved_model_dir={}".format(args.output_dir / "efficientdet-d0_saved_model")
+    ], check=True)
+
+if __name__ == '__main__':
+    main()
diff --git a/models/public/efficientdet-d1-tf/accuracy-check.yml b/models/public/efficientdet-d1-tf/accuracy-check.yml
@@ -0,0 +1,23 @@
+models:
+  - name: efficientdet-d1-tf
+    launchers:
+      - framework: dlsdk
+        adapter: ssd
+    datasets:
+      - name: ms_coco_detection_90_class_without_backgound
+        preprocessing:
+          - type: resize
+            aspect_ratio_scale: fit_to_window
+            size: 640
+          - type: padding
+            size: 640
+            pad_type: right_bottom
+
+        postprocessing:
+          - type: faster_rcnn_postprocessing_resize
+            size: 640
+          - type: shift_labels
+            offset: 1
+
+        metrics:
+          - type: coco_precision
diff --git a/models/public/efficientdet-d1-tf/efficientdet-d1-tf.md b/models/public/efficientdet-d1-tf/efficientdet-d1-tf.md
@@ -0,0 +1,86 @@
+# efficientdet-d1-tf
+
+## Use Case and High-Level Description
+
+The "efficientdet-d1" model is one of the [EfficientDet](https://arxiv.org/abs/1911.09070)
+models  designed to perform object detection. This model was pretrained in TensorFlow*.
+All the EfficientDet models have been pretrained on the MSCOCO* image database.
+For details about this family of models, check out the Google AutoML [repository]
+(https://github.com/google/automl/tree/master/efficientdet).
+
+## Example
+
+## Specification
+
+| Metric            | Value           |
+|-------------------|-----------------|
+| Type              | Object detection|
+| GFLOPs            | 0.819           |
+| MParams           | 5.268           |
+| Source framework  | TensorFlow\*    |
+
+## Accuracy
+
+| Metric | Converted model |
+| ------ | --------------- |
+| coco_precision | 37.54%|
+
+## Performance
+
+## Input
+
+### Original Model
+
+Image, name - `convert_image/Cast`,  shape - `[1x640x640x3]`, format is `[BxHxWxC]`, where:
+
+- `B` - batch size
+- `H` - height
+- `W` - width
+- `C` - channel
+
+Channel order is `RGB`.
+
+### Converted Model
+
+Image, name - `convert_image/Cast/placeholder_port_0`,  shape - `[1x3x640x640]`, format is `[BxCxHxW]`, where:
+
+- `B` - batch size
+- `C` - channel
+- `H` - height
+- `W` - width
+
+Channel order is `BGR`.
+
+## Output
+
+### Original Model
+
+The array of summary detection information, name: `detections`, shape: [1, 7, N], where N is the number of detected
+bounding boxes. For each detection, the description has the format:
+[`image_id`, `y_min`, `x_min`, `y_max`, `x_max`, `confidence`, `label`],
+    where:
+
+    - `image_id` - ID of the image in the batch
+    - (`x_min`, `y_min`) - coordinates of the top left bounding box corner
+    - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner
+    - `confidence` - confidence for the predicted class
+    - `label` - predicted class ID, starting from 1
+
+### Converted Model
+
+The array of summary detection information, name: `detections`, shape: [1, 1, N, 7], where N is the number of detected
+bounding boxes. For each detection, the description has the format:
+[`image_id`, `label`, `conf`, `x_min`, `y_min`, `x_max`, `y_max`],
+    where:
+
+    - `image_id` - ID of the image in the batch
+    - `label` - predicted class ID, starting from 0
+    - `conf` - confidence for the predicted class
+    - (`x_min`, `y_min`) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])
+    - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner  (coordinates stored in normalized format, in range [0, 1])
+
+## Legal Information
+
+The original model is distributed under the
+[Apache License, Version 2.0](https://raw.githubusercontent.com/google/automl/master/LICENSE).
+A copy of the license is provided in [APACHE-2.0-TF-AutoML.txt](../licenses/APACHE-2.0-TF-AutoML.txt).