50 lines (38 loc) · 2 KB

PostProcessing Overview

The PostProcessing folder contains scripts for refining and analyzing data captured during robotic experiments. These scripts process raw or intermediate data to ensure quality, consistency, and readiness for model training or further analysis.

Scripts

1. 1_frame_corrector.py

Purpose: Filters frames from captured data based on significant changes in joint angles.
Key Features:
- Detects and selects frames where changes in joint angles exceed a predefined threshold.
- Saves filtered frames and corresponding depth data for further processing.
Output: A filtered dataset and image files.

2. 2_yolo_world.py

Purpose: Detects objects in experimental images using YOLOWorld object detection.
Key Features:
- Identifies predefined objects (e.g., "red ball", "green box") in each frame.
- Annotates datasets with the detected object's position (origin_x, origin_y, goal_x, goal_y).
Output: Updated datasets with object positions.

3. 3_interpolate.py

Purpose: Interpolates missing or inconsistent values in the dataset.
Key Features:
- Fills gaps in the origin_x, origin_y, goal_x, and goal_y columns using linear interpolation.
- Ensures smooth data continuity for downstream tasks.
Output: A cleaned and interpolated dataset.

4. 4_object_depth.py

Purpose: Adds depth (z) information to detected objects.
Key Features:
- Processes depth data to compute origin_z and goal_z values.
- Combines 2D positional data with depth for 3D representation.
Output: Datasets with added origin_z and goal_z columns.

5. 5_is_grasp.py

Purpose: Labels data with the is_grasped status.
Key Features:
- Detects when the robotic arm grasps an object based on deviation from baseline positions.
- Adds a binary column is_grasped to the dataset for training models.
Output: Datasets annotated with grasping status.