Merge pull request #82 from PUTvision/cars_detection_yolo_tutorial

Cars detection yolo tutorial
PUTvision · Mar 10, 2023 · 6d1d86d · 6d1d86d
2 parents df37d65 + 00f65e7
commit 6d1d86d
Show file tree

Hide file tree

Showing 11 changed files with 554 additions and 17 deletions.
diff --git a/docs/source/conf.py b/docs/source/conf.py
@@ -16,11 +16,11 @@
 import os
 import re
 import sys
-sys.path.append(os.path.abspath('../../plugin/'))
+sys.path.append(os.path.abspath('../../src/'))
 
 # -- Project information -----------------------------------------------------
 
-metadata_file_path = os.path.join('..', '..', 'plugin', 'deepness', 'metadata.txt')
+metadata_file_path = os.path.join('..', '..', 'src', 'deepness', 'metadata.txt')
 metadata_file_path = os.path.abspath(metadata_file_path)
 with open(metadata_file_path, 'rt') as file:
     file_content = file.read()

diff --git a/docs/source/creators/creators_tutorial.rst b/docs/source/creators/creators_tutorial.rst
@@ -0,0 +1,24 @@
+Model creation tutorial
+=======================
+
+
+=========
+Detection
+=========
+
+For one of models in our zoo - specifically for cars detection on aerial images - a complete tutorial is provided in a jupyter notebook:
+
+  .. code-block::
+
+        ./tutorials/detection/cars_yolov7/car_detection__prepare_and_train.ipynb
+
+
+The notebook covers:
+ * downloading yolov7 repository
+ * downloading the training dataset
+ * preparing training data and labels in yolov7 format
+ * running th training and testing
+ * conversion to ONNX model
+ * adding default parameters for Deepness plugin
+
+Example model inference can be found in the :code:`Examples` section.
diff --git a/docs/source/example/example_detection_cars_yolov7.rst b/docs/source/example/example_detection_cars_yolov7.rst
@@ -0,0 +1,46 @@
+YOLOv7 cars detection
+===================================
+
+The following example shows how to use the YOLOv7 model for cars (and other vehicles) detection in aerial or satellite images.
+
+=======
+Dataset
+=======
+
+The example is based on the `ITCVD cars detection dataset <https://arxiv.org/pdf/1801.07339.pdf>`_. It provides aerial images with 10 cm/px resolution. Annotation bounding boxes for the cars are provided.
+
+=========================
+Training tutorial
+=========================
+
+The entire training process has been gathered in a tutorial notebook in jupyter notebook:
+
+
+  .. code-block::
+
+        ./tutorials/detection/cars_yolov7/car_detection__prepare_and_train.ipynb
+
+
+==================
+Example inference
+==================
+
+Run QGIS, next add "Poznan 2022 aerial" map using :code:`QuickMapServices` plugin.
+
+Alternatively you can use any other aerial or satellite map with resolution of at least 10 cm/pixel
+
+.. image:: ../images/cars_near_poznan_university_of_technology_on_ortophoto__zoom_in.webp
+
+Then run our plugin and set parameters like in the screenshot below. You can find the pre-trained onnx model at :code:`https://chmura.put.poznan.pl/s/vgOeUN4H4tGsrGm`. Push the Run button to start processing.
+
+.. image:: ../images/cars_near_poznan_university_of_technology_on_ortophoto.webp
+
+
+Another inference on random street in Poznan:
+
+    .. image:: ../images/cars_on_ransom_street_in_poznan.webp
+
+
+And output mask for an Grunwald district in Poznan:
+
+    .. image:: ../images/ecars_in_poznan_grunwald_district.webp
diff --git a/docs/source/images/cars_in_poznan_grunwald_district.webp b/docs/source/images/cars_in_poznan_grunwald_district.webp
diff --git a/docs/source/images/cars_near_poznan_university_of_technology_on_ortophoto.webp b/docs/source/images/cars_near_poznan_university_of_technology_on_ortophoto.webp
diff --git a/docs/source/images/cars_near_poznan_university_of_technology_on_ortophoto__zoom_in.webp b/docs/source/images/cars_near_poznan_university_of_technology_on_ortophoto__zoom_in.webp
diff --git a/docs/source/images/cars_on_ransom_street_in_poznan.webp b/docs/source/images/cars_on_ransom_street_in_poznan.webp
diff --git a/docs/source/index.rst b/docs/source/index.rst
@@ -38,6 +38,7 @@ Home
    :maxdepth: 1
    :caption: Examples
 
+   example/example_detection_cars_yolov7
    example/example_segmentation_landcover
    example/example_detection_planes_yolov7
    example/example_detection_oils_yolov5
@@ -48,6 +49,7 @@ Home
    :caption: For Model Creators
 
    creators/creators_description_classes
+   creators/creators_tutorial
    creators/creators_export_training_data_tool
    creators/creators_example_onnx_model
    creators/creators_add_metadata_to_model

diff --git a/docs/source/main/model_zoo/MODEL_ZOO.md b/docs/source/main/model_zoo/MODEL_ZOO.md
@@ -4,25 +4,26 @@ The [Model ZOO](https://chmura.put.poznan.pl/s/2pJk4izRurzQwu3) is a collection
 
 ## Segmentation models
 
-| Model name                                                                         | Input size | CM/PX | Description | Example image |
-|------------------------------------------------------------------------------------|---|---|---|---------------------------------------------------------|
-| [Corn Field Damage Segmentation](https://chmura.put.poznan.pl/s/abWFTVYSDIcncWs) | 512 | 3 | [PUT Vision](https://putvision.github.io/) model for Corn Field Damage Segmentation created on own dataset labeled by experts. We used the classical UNet++ model. It generates 3 outputs: healthy crop, damaged crop, and out-of-field area. | [Image](https://chmura.put.poznan.pl/s/i5WVmcfqPNdBTAQ) | 
-| [Land Cover Segmentation](https://chmura.put.poznan.pl/s/PnAFJw27uneROkV)          | 512 | 40 | The model is trained on the [LandCover.ai dataset](https://landcover.ai.linuxpolska.com/). It provides satellite images with 25 cm/px and 50 cm/px resolution. Annotation masks for the following classes are provided for the images: building (1), woodland (2), water(3), road(4). We use `DeepLabV3+` model with `tu-semnasnet_100` backend and `FocalDice` as a loss function. | [Image](https://chmura.put.poznan.pl/s/Xa29vnieNQTvSt5) |
-| [Roads Segmentation](https://chmura.put.poznan.pl/s/y6S3CmodPy1fYYz)            | 512 | 21 | The model segments the Google Earth satellite images into 'road' and 'not-road' classes. Model works best on wide car roads, crossroads and roundabouts.  | [Image](https://chmura.put.poznan.pl/s/rln6mpbjpsXWpKg) |
+| Model                                                                            | Input size | CM/PX | Description                                                                                                                                                                                                                                                                                                                                                                         | Example image |
+|----------------------------------------------------------------------------------|------------|-------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------|
+| [Corn Field Damage Segmentation](https://chmura.put.poznan.pl/s/abWFTVYSDIcncWs) | 512        | 3     | [PUT Vision](https://putvision.github.io/) model for Corn Field Damage Segmentation created on own dataset labeled by experts. We used the classical UNet++ model. It generates 3 outputs: healthy crop, damaged crop, and out-of-field area.                                                                                                                                       | [Image](https://chmura.put.poznan.pl/s/i5WVmcfqPNdBTAQ) | 
+| [Land Cover Segmentation](https://chmura.put.poznan.pl/s/PnAFJw27uneROkV)        | 512        | 40    | The model is trained on the [LandCover.ai dataset](https://landcover.ai.linuxpolska.com/). It provides satellite images with 25 cm/px and 50 cm/px resolution. Annotation masks for the following classes are provided for the images: building (1), woodland (2), water(3), road(4). We use `DeepLabV3+` model with `tu-semnasnet_100` backend and `FocalDice` as a loss function. | [Image](https://chmura.put.poznan.pl/s/Xa29vnieNQTvSt5) |
+| [Roads Segmentation](https://chmura.put.poznan.pl/s/y6S3CmodPy1fYYz)             | 512        | 21    | The model segments the Google Earth satellite images into 'road' and 'not-road' classes. Model works best on wide car roads, crossroads and roundabouts.                                                                                                                                                                                                                            | [Image](https://chmura.put.poznan.pl/s/rln6mpbjpsXWpKg) |
 
 ## Regression models
 
-| Model name | Input size | CM/PX | Description | Example image |
-|---|---|---|---|---|
-|  |  |  |  |  |
-|  |  |  |  |  |
+| Model   | Input size | CM/PX | Description | Example image |
+|---------|---|---|---|---|
+|         |  |  |  |  |
+|         |  |  |  |  |
 
 ## Object detection models
 
-| Model name | Input size | CM/PX | Description | Example image |
-|---|---|---|---|---|
-| [Airbus Planes Detection](https://chmura.put.poznan.pl/s/bBIJ5FDPgyQvJ49) | 256 | 70 | YOLOv7 tiny model for object detection on satellite images. Based on the [Airbus Aircraft Detection dataset](https://www.kaggle.com/datasets/airbusgeo/airbus-aircrafts-sample-dataset). | [Image](https://chmura.put.poznan.pl/s/VfLmcWhvWf0UJfI) |
-| [Airbus Oil Storage Detection](https://chmura.put.poznan.pl/s/gMundpKsYUC7sNb) | 512 | 150 | YOLOv5-m model for object detection on satellite images. Based on the [Airbus Oil Storage Detection dataset](https://www.kaggle.com/datasets/airbusgeo/airbus-oil-storage-detection-dataset). | [Image](https://chmura.put.poznan.pl/s/T3pwaKlbFDBB2C3) |
+| Model                                                                          | Input size | CM/PX | Description                                                                                                                                                                                   | Example image                                           |
+|--------------------------------------------------------------------------------|------------|-------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------|
+| [Airbus Planes Detection](https://chmura.put.poznan.pl/s/bBIJ5FDPgyQvJ49)      | 256        | 70    | YOLOv7 tiny model for object detection on satellite images. Based on the [Airbus Aircraft Detection dataset](https://www.kaggle.com/datasets/airbusgeo/airbus-aircrafts-sample-dataset).      | [Image](https://chmura.put.poznan.pl/s/VfLmcWhvWf0UJfI) |
+| [Airbus Oil Storage Detection](https://chmura.put.poznan.pl/s/gMundpKsYUC7sNb) | 512        | 150   | YOLOv5-m model for object detection on satellite images. Based on the [Airbus Oil Storage Detection dataset](https://www.kaggle.com/datasets/airbusgeo/airbus-oil-storage-detection-dataset). | [Image](https://chmura.put.poznan.pl/s/T3pwaKlbFDBB2C3) |
+| [Aerial Cars Detection](https://chmura.put.poznan.pl/s/vgOeUN4H4tGsrGm)        | 640        | 10    | YOLOv7-m model for cars detection on aerial images. Based on the [ITCVD](https://arxiv.org/pdf/1801.07339.pdf).                                                                               | [Image](https://chmura.put.poznan.pl/s/cPzw1mkXlprSUIJ) |
 
 ## Contributing
 

diff --git a/src/deepness/metadata.txt b/src/deepness/metadata.txt
@@ -6,7 +6,7 @@
 name=Deepness: Deep Neural Remote Sensing
 qgisMinimumVersion=3.22
 description=Inference of deep neural network models (ONNX) for segmentation, detection and regression
-version=0.4.1
+version=0.5.0
 author=PUT Vision
 [email protected]
 
@@ -17,7 +17,7 @@ about=
      - limiting processing range to predefined area (visible part or area defined by vector layer polygons)
      - common types of models are supported: segmentation, regression, detection
      - integration with layers (both for input data and model output layers). Once an output layer is created, it can be saved as a file manually
-     - model ZOO under development (planes detection on Bing Aerial, Corn field damage, Oil Storage tanks detection, ...)
+     - model ZOO under development (planes detection on Bing Aerial, Corn field damage, Oil Storage tanks detection, cars detection, ...)
      - training data Export Tool - exporting raster and mask as small tiles
      - parametrization of the processing for advanced users (spatial resolution, overlap, postprocessing)
     Plugin requires external python packages to be installed. After the first plugin startup, a Dialog will show, to assist in this process. Please visit plugin the documentation for details.