Add GUI for YAML Configuration, Training, and Prediction in AOC Fruit Detector

README.md

@@ -96,6 +96,38 @@ The **config** folder contains two parameter files for specifying system charact
 
 Please note that annotated images and JSON annotation files are saved to the location where the package is installed. Check the **install** folder of the repository to find the annotation outputs.
 
+## 🖥️ Using the Graphical User Interface (GUI)
+
+AOC Fruit Detector includes a visual GUI for editing YAML configurations and launching predictions or training tasks without directly modifying YAML files.
+
+### Features
+
+- Edit all YAML parameters in a structured, visual layout
+- Select between ROS and non-ROS prediction modes
+- Launch model training or prediction with a button
+- Tooltips available for every parameter
+- Required fields are marked with a red `*`
+- Shows result statistics and prediction output paths
+
+### How to run the GUI
+
+From inside the container:
+
+```bash
+cd fruit_detector_ws/
+python src/aoc_fruit_detector/gui_predic_trainer/main.py
+```
+
+This will launch the GUI.
+
+### GUI Screenshot
+
+Make sure the following image exists in your repo at `docs/gui_screenshot.png`:
+
+![GUI screenshot](./gui_predic_trainer/assets/gui_screenshot.png)
+
+---
+
 ## Wiki
 
 To get more information about the AOC Fruit Detector, please refer to [Wiki page][wiki_page]

config/non_ros_params.yaml

@@ -6,9 +6,9 @@ datasets:
   train_dataset_name: 'aoc_train_dataset'
   test_dataset_name: 'aoc_test_dataset'
   validation_dataset_name: 'aoc_validation_dataset'
-  dataset_train_annotation_url: 'https://lncn.ac/aocanntrain' 
+  dataset_train_annotation_url: 'https://lncn.ac/aocanntrain'
   dataset_train_images_url: 'https://lncn.ac/aocdatatrain'
-  dataset_test_annotation_url: 'https://lncn.ac/aocanntest' 
+  dataset_test_annotation_url: 'https://lncn.ac/aocanntest'
   dataset_test_images_url: 'https://lncn.ac/aocdatatest'
 files:
   # pretrained model used as a training base model, if set as empty, the config file will use imagenet trained model as base.
@@ -26,11 +26,11 @@ files:
   train_catalog_url: 'https://lncn.ac/aoccat'
 directories:
   train_image_dir: './data/strawberry_dataset/train/'
-  test_image_dir: './data/bag/rgbd/' #'./data/strawberry_dataset/test/'
+  test_image_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/bag/rgbd' #'./data/strawberry_dataset/test/'
   validation_image_dir: './data/tomato_dataset/val/'
   training_output_dir: './data/training_output/'
-  prediction_output_dir: './data/prediction_output/test_images/'
-  prediction_json_dir: './data/annotations/predicted/' 
+  prediction_output_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data'
+  prediction_json_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data'
 training:
   epochs: 40000
   number_of_classes: 2
@@ -44,9 +44,8 @@ settings:
   show_orientation: true
   fruit_type: 'strawberry' # currently supported for "strawberry" or "tomato"
   validation_period: 500 # Smaller validation will increase training time. The value is set to have 100 validation during training
-  confidence_threshold: 0.9 # removes annotations with confidence score lower than this value
-  filename_patterns: 
-    rgb: "image" 
-    depth: "depth" # filename patterns for rgb and depth images
+  confidence_threshold: 0.5 # removes annotations with confidence score lower than this value
+  filename_patterns:
+    rgb: "_rgb"
+    depth: "_depth" # filename patterns for rgb and depth images
 
-

config/ros_params.yaml

@@ -9,4 +9,4 @@ aoc_fruit_detector:
     verbose: [True, True, True, True, True] # To draw [centroid, bbox, mask, coordinate frames, text] annotations on image
     pub_verbose: True # To publish annotated image
     pub_markers: True # To publish RViz markers
-    use_ros: True # To subscribe RGBD data and publish output from/on ROS2 framework
+    use_ros: False # To subscribe RGBD data and publish output from/on ROS2 framework

config/training_parameters.yaml

@@ -0,0 +1,51 @@
+#Call program with "python -O" O=Optimize directive for non debug mode. (it is not a zero :))
+#When executed in Optimized mode, the program will neither output predicted images with bounding box/mask overlays nor it will write a respective json file. Moreover, the debug traces will not print at error event. In this mode json will be output in a variable.
+
+# UZ: After training, the newly trained model should be place in the model folder, it should be renamed to aoc_model.pth 
+datasets:
+  train_dataset_name: 'flowers_train'
+  test_dataset_name: 'flowers_test'
+  validation_dataset_name: 'flowers_val'
+  dataset_train_annotation_url: '' #'https://lncn.ac/aocanntrain' 
+  dataset_train_images_url: '' #'https://lncn.ac/aocdatatrain'
+  dataset_test_annotation_url: '' #'https://lncn.ac/aocanntest' 
+  dataset_test_images_url: '' #'https://lncn.ac/aocdatatest'
+files:
+  # pretrained model used as a training base model, if set as empty, the config file will use imagenet trained model as base.
+  pretrained_model_file: ''
+  # model_file: './model/aoc_tomato_ripeness_151_40k.pth'
+  model_file: './model/aoc_flowers.pth' #'./model/aoc_strawberry_class_ripeness.pth' #'./model/aoc_model.pth'
+  config_file: 'COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml'
+  test_metadata_catalog_file: './data/dataset_catalogs/flowers_test_metadata.pkl' #'./data/dataset_catalogs/tom_test_metadata_catalog.pkl'
+  train_dataset_catalog_file: './data/dataset_catalogs/flowers_train_metadata.pkl' #'./data/dataset_catalogs/tom_train_dataset_catalog.pkl'
+  train_annotation_file: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/annotations/train_annotations.json' #'./data/tomato_dataset/train/annotations/ripeness_class_annotations.json'
+  test_annotation_file: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/annotations/test_annotations.json' #'./data/tomato_dataset/test/annotations/ripeness_class_annotations.json'
+  validation_annotation_file: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/annotations/val_annotations.json' #'./data/tomato_dataset/val/annotations/ripeness_class_annotations.json'
+  model_url: '' #'https://lncn.ac/aocmodel'
+  meta_catalog_url: '' #'https://lncn.ac/aocmeta'
+  train_catalog_url: '' #'https://lncn.ac/aoccat'
+directories:
+  train_image_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/train/' #'./data/strawberry_dataset/train/'
+  test_image_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/test/'  #'./data/strawberry_dataset/test/'
+  validation_image_dir: '/home/ros/fruit_detector_ws/src/aoc_fruit_detector/scripts/data/flowers_dataset/val/' #'./data/tomato_dataset/val/'
+  training_output_dir: './data/training_output/'
+  prediction_output_dir: './data/prediction_output/test_images/'
+  prediction_json_dir: './data/annotations/predicted/'
+training:
+  epochs: 40000
+  number_of_classes: 1
+  optimizer: 'SGD'
+  learning_rate: 0.0025
+settings:
+  download_assets: false # if assets such as model and datasets should be downloaded
+  rename_pred_images: false #rename the predicted images in img_000001.png like format
+  segm_masks: false
+  bbox: true
+  show_orientation: false
+  fruit_type: '' # currently supported for "strawberry" or "tomato"
+  validation_period: 500 # Smaller validation will increase training time. The value is set to have 100 validation during training
+  confidence_threshold: 0.5 # removes annotations with confidence score lower than this value
+  filename_patterns:
+    rgb: "_rgb"
+    depth: "_depth" # filename patterns for rgb and depth images
+

gui_predic_trainer/config_paths.py

@@ -0,0 +1,13 @@
+# ============================================================
+# File: config_paths.py
+# Author: Yael Vicente
+# Date: 2025-May-06
+# Description:
+#   Centralized configuration paths used across the GUI.
+#   Defines static paths to parameter files for training and prediction.
+# ============================================================
+
+# Paths to configuration files
+TRAINING_CONFIG_PATH = "/home/ros/fruit_detector_ws/src/aoc_fruit_detector/config/training_parameters.yaml"
+ROS_PARAMS_PATH = "/home/ros/fruit_detector_ws/src/aoc_fruit_detector/config/ros_params.yaml"
+NON_ROS_PARAMS_PATH = "/home/ros/fruit_detector_ws/src/aoc_fruit_detector/config/non_ros_params.yaml"

gui_predic_trainer/constants.py

@@ -0,0 +1,73 @@
+# ============================================================
+# File: constants.py
+# Author: Yael Vicente
+# Date: 2025-May-06
+# Description:
+#   Dictionary containing formal English descriptions for YAML configuration
+#   fields used in the AOC Fruit Detector GUI. These descriptions are
+#   displayed as tooltips to assist the user during configuration.
+# ============================================================
+
+FIELD_DESCRIPTIONS = {
+
+    # ROS-specific parameters (aoc_fruit_detector.ros__parameters.*)
+    "aoc_fruit_detector.ros__parameters.min_depth": "Minimum depth threshold (in meters) for considering objects during detection.",
+    "aoc_fruit_detector.ros__parameters.max_depth": "Maximum depth threshold (in meters) to validate detected objects.",
+    "aoc_fruit_detector.ros__parameters.constant_depth_value": "Fixed depth value to be used when depth data is not available.",
+    "aoc_fruit_detector.ros__parameters.fruit_type": "Specifies the type of fruit to detect within the ROS pipeline. Supported values: 'strawberry', 'tomato'.",
+    "aoc_fruit_detector.ros__parameters.pose3d_frame": "TF frame used to publish the 3D pose of detected fruits.",
+    "aoc_fruit_detector.ros__parameters.pose3d_tf": "Determines whether pose transformation between frames is required (True = apply transformation).",
+    "aoc_fruit_detector.ros__parameters.verbose": "List of boolean flags to enable visual annotations: [centroid, bounding box, mask, coordinate frame, text].",
+    "aoc_fruit_detector.ros__parameters.pub_verbose": "Enables publishing of annotated prediction images on a ROS topic.",
+    "aoc_fruit_detector.ros__parameters.pub_markers": "Enables publishing of RViz markers for 3D fruit pose visualization.",
+    "aoc_fruit_detector.ros__parameters.use_ros": "If enabled, the model will run within the ROS2 framework using topics for I/O.",
+
+    # Dataset-related fields
+    "datasets.train_dataset_name": "Name of the training dataset. Use the same name as the dataset used to train the selected model. Required.",
+    "datasets.test_dataset_name": "Name of the testing dataset. Use the same name as the dataset used to train the selected model. Required.",
+    "datasets.validation_dataset_name": "Name of the validation dataset. Use the same name as the dataset used to train the selected model. Required.",
+    "datasets.dataset_train_annotation_url": "URL for the training dataset's annotation file. Required if 'download_assets' is enabled.",
+    "datasets.dataset_train_images_url": "URL for the training dataset's image folder. Required if 'download_assets' is enabled.",
+    "datasets.dataset_test_annotation_url": "URL for the test dataset's annotation file. Required if 'download_assets' is enabled.",
+    "datasets.dataset_test_images_url": "URL for the test dataset's image folder. Required if 'download_assets' is enabled.",
+
+    # File paths
+    "files.pretrained_model_file": "Path to a pretrained model used for fine-tuning. Optional.",
+    "files.model_file": "Path to the trained model to be used. If training a new model, specify the desired output filename.",
+    "files.config_file": "Path to the Detectron2 configuration file (.yaml). Required. Do not change unless using a custom config.",
+    "files.test_metadata_catalog_file": "Path to the metadata catalog for testing. Optional, unless using a custom metadata structure.",
+    "files.train_dataset_catalog_file": "Path to the training dataset catalog. Optional, unless using a custom dataset setup.",
+    "files.train_annotation_file": "Path to the COCO-format annotation file for training. Required if 'download_assets' is disabled.",
+    "files.test_annotation_file": "Path to the COCO-format annotation file for testing. Required if 'download_assets' is disabled.",
+    "files.validation_annotation_file": "Path to the COCO-format annotation file for validation. Optional but recommended.",
+    "files.model_url": "URL to download model weights. Required if 'download_assets' is enabled and no local model is provided.",
+    "files.meta_catalog_url": "URL to download the metadata catalog. Required if 'download_assets' is enabled and no local catalog is provided.",
+    "files.train_catalog_url": "URL to download the training dataset catalog. Required if 'download_assets' is enabled and no local file is provided.",
+
+    # Directories
+    "directories.train_image_dir": "Directory containing images for training. Required for both training and evaluation.",
+    "directories.test_image_dir": "Directory containing images to be used for prediction. Required for inference mode.",
+    "directories.validation_image_dir": "Directory containing validation images. Optional unless validation is enabled.",
+    "directories.training_output_dir": "Directory where training logs and outputs will be saved.",
+    "directories.prediction_output_dir": "Directory where predicted images will be saved. Required in both training and prediction modes.",
+    "directories.prediction_json_dir": "Directory where prediction JSON files will be stored. Required in both training and prediction modes.",
+
+    # Training parameters
+    "training.epochs": "Number of training epochs to run. Must be a positive integer.",
+    "training.batch_size": "Number of samples per batch during training. Typical values range from 2 to 32.",
+    "training.number_of_classes": "Total number of classes the model should predict. Required.",
+    "training.optimizer": "Optimizer algorithm to use for training (e.g., 'SGD', 'Adam').",
+    "training.learning_rate": "Learning rate used by the optimizer. Typical values range from 0.001 to 0.01.",
+
+    # Model settings
+    "settings.download_assets": "Set to True to enable downloading of datasets and model files. Set to False if all assets are already available locally.",
+    "settings.rename_pred_images": "Set to True to rename predicted image files using a numbered pattern (e.g., img_000001.png).",
+    "settings.segm_masks": "Set to True to enable segmentation mask predictions. Set to False to disable.",
+    "settings.bbox": "Set to True to enable bounding box predictions. Set to False to disable.",
+    "settings.show_orientation": "Set to True to enable orientation estimation for detections. Only supported for strawberries and tomatoes.",
+    "settings.fruit_type": "Name of the fruit to be detected. Supported values: 'strawberry', 'tomato'. Leave empty if using a model trained on a different fruit.",
+    "settings.validation_period": "Applies during training. Smaller values increase validation frequency, which may increase training time.",
+    "settings.confidence_threshold": "Removes predictions with confidence scores below this threshold.",
+    "settings.filename_patterns.rgb": "Pattern used to identify RGB images by filename.",
+    "settings.filename_patterns.depth": "Pattern used to identify depth images by filename."
+}