Skip to content

Latest commit

 

History

History
31 lines (28 loc) · 1.99 KB

RunWithoutROS.md

File metadata and controls

31 lines (28 loc) · 1.99 KB
Raw

Video Annotations and Data Collection with IST-ROS

This document provides a guide on using IST_ROS for video annotations and data collection, using a video file in mp4 format as input. The provided script allows users to select target objects in the input video and outputs an annotated video with the selected objects segmented in each frame.

Basic example

To perform video annotation, run the following command:

python src/segment_and_track.py --input_path ./data/inputs/input_video.mp4 --output_path ./data/outputs/videos/output_video.mp4

Arguments

  • --input_path: Path to the input video.
  • --output_path: Path to the output video.
  • --point_mode: Track the targets using points.
  • --disable_mask_mode: Disable the mask mode for visualization.
  • --save_all_mode: Save all visualized frames and/or point positions. The saved data depends on the selected visualizing mode.
  • --save_all_path: Path to save data when --save_all_mode is enabled.

Workflow

  1. Run the script with the appropriate arguments.
  2. A tkinter GUI window will appear.
  3. Select one target at a time using the following operations:
    • Draw a bounding box around the object by clicking and dragging the mouse.
    • Right-click on the object to specify it as foreground or background. You can change the mode using bottun.
  4. Click the "Generate Masks" button to display and register the masked image.
    • If you are not satisfied with the results, use the "Reselect" or "Clear All" button to remove them.
  5. Repeat steps 3-4 for each object you want to track.
  6. After selecting all targets, click the "Start Tracking" button to initiate the video annotation.

For detailed instructions on using the GUI, please refer to GUI.md.

Outputs

The script will process the input video and save the output video with the tracked objects at the specified output path (--output_path). If --save_all_mode is enabled, the script will also save all visualized frames and/or mask positions at the specified path (--save_all_path).