# cis-581-final-project-video-stiching-detection

**Repository Path**: regiontech/cis-581-final-project-video-stiching-detection

## Basic Information

- **Project Name**: cis-581-final-project-video-stiching-detection
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-11-18
- **Last Updated**: 2024-11-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Video Stitching and Detection
Repository for our CIS581 Final Project with Professor Jianbo Shi, Fall 2020.

Srisa Changolkar, Celestina Saven, Vinay Senthil, Kevin Sun (Team 02)

### Description of Code and Layout
* Report: `CIS_581_Final_Project_Report.pdf`
* Video Presentation Slides: `CIS_581_Final_Project_Video_Presentation.pptx`
* Video Presentation: Google Drive link at `CIS_581_Final_Project_Video_Link.txt` or can be directly accessed here: https://drive.google.com/file/d/1XW2AXkco0IthfyUPP2rPwdmLc54nlFfE/view?usp=sharing
* GitHub Repo: Link at `github-repo-link.txt` or can be directly accessed here: https://github.com/ksun0/cis-581-final-project-video-stiching-detection


The root directory has the `requirements.txt` for required libraries and packages to install. `video-stitching-detection.py` contains code to stitch videos into a singular image, as well perform object detection on it. The accompanying notebook `video-sitching.ipynb` contains intermediate results, explanations, and visualizations for the video stitching segment. The `object_detection.ipynb` contains the process for object detection.

The `data/` folder contains all the input videos, as well as intermediate results and final output images. For example, stitching `data/desk-left-right.mp4` puts intermediate results in `data/desk-left-right/`, and outputs the stiched result in `data/desk-left-right.png`. The training data for object detection transfer learning is under `data/training_data/`, including training images and the manual annotations from makesense.ai at `data/annotations.csv`. The model is outputted from `object_detection.ipynb` into `data/object_detection_model.pt`, and finally, object detection will output the stitched result with object labels and rectangles in `data/desk-left-right-detected.png`.

Finally, the `helper/` folder is for Pytorch. `img/` is for documentation images.

### Installation and Setting Up
1. `pip install -r requirements.txt` in a `Python3.6` virtual environment
- Note: some packages might have to be manually installed/removed depending on Operating System
- Pay special attention to scikit-image and pytorch (torch and torchvision)
2. Download `data/object_detection_model.pt` from: https://drive.google.com/file/d/1KKbfRWbitGK83OS16DuSuMgf_4L6aJDj/view?usp=sharing
3. `python video-stitching-detection.py`
- this python file will run four stitches from the 4 commands at the bottom, e.g. `video_stitch_optical_flow('data/desk-left-right.mp4', 'data/desk-left-right', 0, 15, 390, 'data/desk-left-right.png')`
  - the first argument is the path to the video file to stitch together
  - the second arguemnt is the path to the folder to store/generate the static frames to stitch together
  - the third argument is the rotation (e.g. if drone is flying left to right, rotation is 0, but if the drone is flying bottom to up, rotation is 270)
  - the fourth argument is the number of frames so skip for each static frame to stitch
  - the fifth argument is the last frame to process
  - the sixth argument is the output path of the final stitched image.
- Finally, object detection code will run on `data/desk-left-right.png`. Since we trained our model on desk setups, outputting the detected results in `data/left-right-detected.png`.

---

## License
This project is licensed under the MIT License - see the [LICENSE.md](LICENSE.md) file for details

## Acknowledgments
* CIS581 Instructors, TA's, and Piazza