This repository contains code for visualizing the trajectories of sequences from the 7Scenes Dataset. The visualization helps in analyzing the camera poses and movement paths across the dataset sequences.
This project is divided into three major visualization parts:
- Single Plot for All Sequences: All the sequences are plotted together on a single 3D plot to provide an overall visualization of the entire dataset.
- Colored Plot for Each Sequence: All sequences are plotted on the same 3D plot, but each sequence is represented with a different color to distinguish between them.
- Separate Plot for Each Sequence: Each sequence is plotted individually on its own 3D plot to provide a detailed trajectory visualization.
The dataset and poses are preprocessed to align and normalize the pose data before visualization.
The 7Scenes Dataset is an RGB-D dataset that contains sequences captured using a handheld Kinect RGB-D camera in different indoor environments, such as offices and rooms. The dataset is primarily used for visual localization tasks and camera pose regression.
The dataset consists of seven different scenes, with each scene containing multiple sequences. Each sequence has images and their corresponding camera pose (in .pose.txt format) describing the camera's 3D position and orientation.
The code provided in this repository allows you to visualize the camera trajectories of the 7Scenes dataset. The following libraries are used:
OpenCV: For reading images.Matplotlib: For plotting 3D trajectories.transforms3d: For quaternion calculations.PyTorch: For managing datasets and data loading.
The main functionalities include:
- Loading and processing the pose data (
load_and_process_posesfunction). - Defining a custom dataset class (
FireDataset) to load images, depth data, and poses. - Visualizing the trajectories using three different methods.
The following Python packages are required:
transforms3dopencv-pythonmatplotlibtorchtorchvisionpickle