Test Results Visualizer

A comprehensive GUI application for visualizing and analyzing computer vision datasets, ground truth annotations, and object detection test results with 6D pose estimation capabilities.

Features

Ground Truths Visualization Tab

• Dataset Processing: Read and process multi-camera RGB-D datasets
• 3D Models Integration: Load and associate 3D object models with scene data
• Ground Truth Visualization: Display RGB images with mask overlays and bounding boxes
• 3D Point Cloud Generation: Create colored point clouds from RGB-D data
• Pose Visualization: View ground truth 6D object poses in 3D space
• Training Data Analysis: Analyze class distributions and dataset statistics
• Mask R-CNN Training Interface: Prepare data for model training

Test Results Visualization Tab

• Test Dataset Processing: Process test scenes for object detection evaluation
• Object Detection Pipeline: Run Mask R-CNN inference on test images
• Detection Consolidation: Merge multi-view detections using clustering algorithms
• Feature Matching: Perform SuperGlue-based feature matching between views
• 6D Pose Estimation: Compute object poses in camera and world coordinates
• Multi-View Navigation: Browse through detection results across different camera views
• 3D Results Visualization: Display detected objects and estimated poses in 3D space

Requirements

Python Dependencies

tkinter
opencv-python (cv2)
numpy
open3d
PIL (Pillow)
json
csv
threading
collections

Hardware Requirements

• GPU support recommended for point cloud processing and inference
• Sufficient RAM for large point cloud visualization
• OpenGL support for 3D visualization

Installation

1. Clone the repository:

git clone [your-repository-url]
cd test-results-visualizer

2. Install required dependencies:

pip install opencv-python numpy open3d pillow

3. Ensure you have the required helper modules in the helpers/ directory:

• read_dataset.py
• read_test_dataset.py
• pxl_2_point.py
• process_scene_data.py

Usage

Starting the Application

python ui.py

Ground Truths Tab Workflow

1. Select Dataset Root Folder: Choose the main dataset directory containing RGB images and depth data
2. Select 3D Models Path: Choose directory containing 3D object models (.ply files)
3. Process Dataset: Click "Read Dataset" to generate master CSV with scene annotations
4. Load Image + PCL: Select a row from the CSV table and load corresponding RGB-D data
5. Visualize in 3D: Click "Open 3D Viewer" to see point cloud with ground truth poses

Test Results Tab Workflow

1. Select Dataset Path: Choose test dataset directory
2. Process Test Dataset: Generate test CSV with scene information
3. Load Image + PCL: Select a scene and load RGB-D data
4. Run Detection: Execute the full detection and pose estimation pipeline
5. Navigate Results: Use arrow buttons to browse through detection results and feature correspondences
6. Visualize in 3D: View point cloud with estimated object poses

Method

This application implements a multi-view object detection and 6D pose estimation pipeline for RGB-D datasets. The method consists of several key stages:

Pipeline Overview

The detection pipeline processes multi-camera RGB-D scenes through the following stages. The system consolidates detections across multiple camera views using spatial clustering and feature-based correspondence.

---

The method leverages:

• Mask R-CNN for initial object detection and segmentation
• Multi-view geometric constraints for detection consolidation
• SuperGlue feature matching for establishing correspondences
• PnP algorithms for 6D pose estimation
• Point cloud registration for pose refinement

File Structure

├── ui.py                          # Main application entry point
├── ground_truths_tab.py           # Ground truth visualization interface
├── test_results_tab.py            # Test results visualization interface
├── helpers/
│   ├── read_dataset.py            # Dataset processing utilities
│   ├── read_test_dataset.py       # Test dataset processing
│   ├── pxl_2_point.py            # Point cloud generation
│   └── process_scene_data.py      # Scene processing and detection pipeline
└── meta_data/                     # Generated CSV files and metadata

Dataset Format

Expected Directory Structure

dataset_root/
├── rgb_cam1/                      # RGB images from camera 1
├── rgb_cam2/                      # RGB images from camera 2
├── rgb_photoneo/                  # RGB images from Photoneo camera
├── depth_cam1/                    # Corresponding depth images
├── depth_cam2/
├── depth_photoneo/
├── scene_camera_cam1.json         # Camera intrinsic parameters
├── scene_camera_cam2.json
└── scene_camera_photoneo.json

Camera Parameter Format

{
  "0": {
    "cam_K": [fx, 0, cx, 0, fy, cy, 0, 0, 1],
    "depth_scale": 1000
  }
}

Features and Controls

Image Visualization

• Zoom Controls: Use +/- buttons to zoom in/out of images
• Pan Navigation: Click and drag to pan around zoomed images
• Multi-View Navigation: Use arrow buttons to switch between camera views
• Mask Overlays: Ground truth masks and detection results overlay on RGB images

3D Visualization

• Point Cloud Display: Colored 3D point clouds from RGB-D data
• Pose Markers: Coordinate frames showing object orientations
• Interactive Navigation: Mouse controls for 3D scene exploration
• Multi-Object Support: Simultaneous visualization of multiple detected objects

Troubleshooting

Common Issues

"No folder selected" Error

• Ensure you've selected both dataset root folder and 3D models path before processing

"Depth image not found" Error

• Check that depth images exist in expected directories (depth_cam1/, depth_photoneo/, etc.)
• Verify image filenames match between RGB and depth directories

Point Cloud Display Issues

• Ensure depth_scale parameter is correct for your dataset
• Check that camera intrinsic parameters are properly formatted
• Verify OpenGL support for 3D visualization

Memory Issues with Large Datasets

• Close 3D viewer windows when not needed
• Process smaller subsets of data for testing
• Ensure sufficient RAM for point cloud processing

Detection Pipeline Errors

• Verify all helper modules are present in helpers/ directory
• Check that required model files are accessible
• Ensure GPU memory is sufficient for inference

Performance Tips

• Use appropriate image scaling for large datasets
• Close unused visualization windows
• Process test scenes in smaller batches
• Enable GPU acceleration when available

Output Files

The application generates several output files in the meta_data/ directory:

• master_dataset_[timestamp].csv: Complete dataset annotations with ground truth poses
• test_dataset.csv: Test scene information for evaluation
• Temporary detection and feature matching results

Contributing

When adding new features:

1. Follow the existing tab-based architecture
2. Implement proper error handling for file I/O operations
3. Add progress indicators for long-running operations
4. Maintain consistent UI styling with existing tabs

License

[Insert your license information here]