Merge pull request openvinotoolkit#1453 from ygnn123/develop

Add vehicle models

models/intel/index.md

@@ -45,6 +45,9 @@ network to detect objects of the same type better.
 | [vehicle-detection-0200](./vehicle-detection-0200/description/vehicle-detection-0200.md)                                                                                            | 0.786                | 1.817      |       |         | X        |       |                |         |
 | [vehicle-detection-0201](./vehicle-detection-0201/description/vehicle-detection-0201.md)                                                                                            | 1.768                | 1.817      |       |         | X        |       |                |         |
 | [vehicle-detection-0202](./vehicle-detection-0202/description/vehicle-detection-0202.md)                                                                                            | 3.143                | 1.817      |       |         | X        |       |                |         |
+| [vehicle-detection-0203](./vehicle-detection-0203/description/vehicle-detection-0203.md)                                                                                            | 112.34               | 24.11      |       |         | X        |       |                |         |
+| [vehicle-detection-0204](./vehicle-detection-0204/description/vehicle-detection-0204.md)                                                                                            | 190.32               | 34.15      |       |         | X        |       |                |         |
+| [vehicle-detection-0205](./vehicle-detection-0205/description/vehicle-detection-0205.md)                                                                                            | 212.22                | 36.56      |       |         | X        |       |                |         |
 | [person-vehicle-bike-detection-crossroad-0078](./person-vehicle-bike-detection-crossroad-0078/description/person-vehicle-bike-detection-crossroad-0078.md)                          | 3.964                | 1.178      |       | X       | X        | X     |                |         |
 | [person-vehicle-bike-detection-crossroad-1016](./person-vehicle-bike-detection-crossroad-1016/description/person-vehicle-bike-detection-crossroad-1016.md)                          | 3.560                | 2.887      |       | X       | X        | X     |                |         |
 | [person-vehicle-bike-detection-crossroad-yolov3-1020](./person-vehicle-bike-detection-crossroad-yolov3-1020/description/person-vehicle-bike-detection-crossroad-yolov3-1020.md)     | 65.984               | 61.922     |       | X       | X        | X     |                |         |

models/intel/vehicle-detection-0203/accuracy-check.yml

@@ -0,0 +1,22 @@
+models:
+  - name: vehicle-detection-0203
+
+    launchers:
+      - framework: dlsdk
+        adapter:
+           type: class_agnostic_detection
+
+    datasets:
+      - name: crossroad_extra_untagged_vehicle_labels_from_1
+        preprocessing:
+          - type: resize
+            dst_width: 1344
+            dst_height: 800
+        postprocessing:
+          - type: resize_prediction_boxes
+            rescale: True 
+          - type: clip_boxes
+            apply_to: prediction
+        metrics:
+          - type: coco_orig_precision
+            include_boundaries: false

models/intel/vehicle-detection-0203/description/vehicle-detection-0203.md

@@ -0,0 +1,49 @@
+# vehicle-detection-0203
+
+## Use Case and High-Level Description
+
+This is a vehicle detector that is based on ResNet18
+backbone with FPN and CascadeRCNN heads. 
+
+## Example
+
+![](./vehicle-detection-0203.png)
+
+## Specification
+
+| Metric                          | Value                                     |
+|---------------------------------|-------------------------------------------|
+| AP @ [ IoU=0.50:0.95 ]          | 0.438 (internal test set)                 |
+| GFlops                          | 112.34                                    |
+| MParams                         | 24.11                                    |
+| Source framework                | PyTorch\*                                 |
+
+Average Precision (AP) is defined as an area under
+the [precision/recall](https://en.wikipedia.org/wiki/Precision_and_recall)
+curve.
+
+## Performance
+
+## Inputs
+
+Name: `input`, shape: [1x3x800x1344] - An input image in the format [BxCxHxW],
+where:
+
+- B - batch size
+- C - number of channels
+- H - image height
+- W - image width
+
+Expected color order is BGR.
+
+## Outputs
+
+The net outputs blob with shape: [1, 1, N, 5], where N is the number of detected
+bounding boxes. Each detection has the format
+  [`x_min`, `y_min`, `x_max`, `y_max`, `conf`], where:
+  - (`x_min`, `y_min`) - coordinates of the top left bounding box corner
+  - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner.
+  - `conf` - confidence for the predicted class
+
+## Legal Information
+[*] Other names and brands may be claimed as the property of others.

models/intel/vehicle-detection-0204/accuracy-check.yml

@@ -0,0 +1,21 @@
+models:
+  - name: vehicle-detection-0204
+
+    launchers:
+      - framework: dlsdk
+        adapter:
+           type: class_agnostic_detection
+    datasets:
+      - name: crossroad_extra_untagged_vehicle_labels_from_1
+        preprocessing:
+          - type: resize
+            dst_width: 1344
+            dst_height: 800
+        postprocessing:
+          - type: resize_prediction_boxes
+            rescale: True 
+          - type: clip_boxes
+            apply_to: prediction
+        metrics:
+          - type: coco_orig_precision
+            include_boundaries: false

models/intel/vehicle-detection-0204/description/vehicle-detection-0204.md

@@ -0,0 +1,49 @@
+# vehicle-detection-0204
+
+## Use Case and High-Level Description
+
+This is a vehicle detector that is based on ResNet34
+backbone with FPN and CascadeRCNN heads. 
+
+## Example
+
+![](./vehicle-detection-0204.png)
+
+## Specification
+
+| Metric                          | Value                                     |
+|---------------------------------|-------------------------------------------|
+| AP @ [ IoU=0.50:0.95 ]          | 0.473 (internal test set)                 |
+| GFlops                          | 190.32                                    |
+| MParams                         | 34.15                                     |
+| Source framework                | PyTorch\*                                 |
+
+Average Precision (AP) is defined as an area under
+the [precision/recall](https://en.wikipedia.org/wiki/Precision_and_recall)
+curve.
+
+## Performance
+
+## Inputs
+
+Name: `input`, shape: [1x3x800x1344] - An input image in the format [BxCxHxW],
+where:
+
+- B - batch size
+- C - number of channels
+- H - image height
+- W - image width
+
+Expected color order is BGR.
+
+## Outputs
+
+The net outputs blob with shape: [1, 1, N, 5], where N is the number of detected
+bounding boxes. Each detection has the format
+  [`x_min`, `y_min`, `x_max`, `y_max`, `conf`], where:
+  - (`x_min`, `y_min`) - coordinates of the top left bounding box corner
+  - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner.
+  - `conf` - confidence for the predicted class
+
+## Legal Information
+[*] Other names and brands may be claimed as the property of others.

models/intel/vehicle-detection-0205/accuracy-check.yml

@@ -0,0 +1,21 @@
+models:
+  - name: vehicle-detection-0205
+
+    launchers:
+      - framework: dlsdk
+        adapter:
+           type: class_agnostic_detection
+    datasets:
+      - name: crossroad_extra_untagged_vehicle_labels_from_1
+        preprocessing:
+          - type: resize
+            dst_width: 1344
+            dst_height: 800
+        postprocessing:
+          - type: resize_prediction_boxes
+            rescale: True 
+          - type: clip_boxes
+            apply_to: prediction
+        metrics:
+          - type: coco_orig_precision
+            include_boundaries: false

models/intel/vehicle-detection-0205/description/vehicle-detection-0205.md

@@ -0,0 +1,49 @@
+# vehicle-detection-0205
+
+## Use Case and High-Level Description
+
+This is a vehicle detector that is based on ResNet50
+backbone with FPN and CascadeRCNN heads. 
+
+## Example
+
+![](./vehicle-detection-0205.png)
+
+## Specification
+
+| Metric                          | Value                                     |
+|---------------------------------|-------------------------------------------|
+| AP @ [ IoU=0.50:0.95 ]          | 0.476                                     |
+| GFlops                          | 212.22                                    |
+| MParams                         | 36.56                                     |
+| Source framework                | PyTorch\*                                 |
+
+Average Precision (AP) is defined as an area under
+the [precision/recall](https://en.wikipedia.org/wiki/Precision_and_recall)
+curve.
+
+## Performance
+
+## Inputs
+
+Name: `input`, shape: [1x3x800x1344] - An input image in the format [BxCxHxW],
+where:
+
+- B - batch size
+- C - number of channels
+- H - image height
+- W - image width
+
+Expected color order is BGR.
+
+## Outputs
+
+The net outputs blob with shape: [1, 1, N, 5], where N is the number of detected
+bounding boxes. Each detection has the format
+  [`x_min`, `y_min`, `x_max`, `y_max`, `conf`], where:
+  - (`x_min`, `y_min`) - coordinates of the top left bounding box corner
+  - (`x_max`, `y_max`) - coordinates of the bottom right bounding box corner.
+  - `conf` - confidence for the predicted class
+
+## Legal Information
+[*] Other names and brands may be claimed as the property of others.

tools/accuracy_checker/configs/vehicle-detection-0203.yml

@@ -0,0 +1 @@
+../../../models/intel/vehicle-detection-0203/accuracy-check.yml

tools/accuracy_checker/configs/vehicle-detection-0204.yml

@@ -0,0 +1 @@
+../../../models/intel/vehicle-detection-0204/accuracy-check.yml

tools/accuracy_checker/configs/vehicle-detection-0205.yml

@@ -0,0 +1 @@
+../../../models/intel/vehicle-detection-0205/accuracy-check.yml

tools/accuracy_checker/dataset_definitions.yml

@@ -414,6 +414,15 @@ datasets:
     annotation: crossroad_extra_untagged_vehicle.pickle
     dataset_meta: crossroad_extra_untagged_vehicle.json
 
+  - name: crossroad_extra_untagged_vehicle_labels_from_1
+    data_source: crossroad_extra_untagged/val_data
+    annotation_conversion:
+      converter: mscoco_detection
+      annotation_file: crossroad_extra_untagged/val_coco_no_bg_vehicle.json
+      has_background: True
+    annotation: crossroad_extra_untagged_vehicle_labels_from_1.pickle
+    dataset_meta: crossroad_extra_untagged_vehicle_labels_from_1.json
+
   - name: pedestrian_and_vehicle_dataset
     data_source: val2017
     annotation_conversion: