wellshcorgi / superres-pano360

SuperRes-Pano360은 입력받은 비디오를 샷으로 분할하고 'SRCNN'을 사용하여 이미지 품질을 향상시킨다. 이미지 스티칭을 하여 360도 파노라마 이미지를 생성합니다.

SuperRes-Pano360

PanoVision360 is a Python-based project that facilitates the creation of 360-degree panoramas from videos. It achieves this by splitting the video into shots, enhancing the image quality using SRCNN (Super-Resolution Convolutional Neural Network), and then stitching the enhanced images together.

Features

• Shot Extraction: The project extracts frames from a video at specified intervals.
• Image Enhancement: Optionally, the extracted frames can be enhanced using SRCNN to improve image quality.
• Panorama Stitching: Finally, the enhanced frames are stitched together to create a seamless 360-degree panorama.

Usage

Test Environment: The project was tested on an Apple MacBook M2 (MacOS), Windows (You Should change directory on source code)

Prerequisites

• Python 3.9
• OpenCV
• TensorFlow
• Pre-train Model Weight Link

Installation

1. Clone the repository:

   git clone https://github.com/WellshCorgi/SuperRes-Pano360.git
   cd SuperRes-Pano360

2. Install dependencies:

   pip install -r requirements.txt

Instructions

1. Place your video file (e.g., input.mp4) in the project directory.

2. Run the script mp4_to_pic.py to extract frames from the video and optionally enhance them:

   python mp4_to_pic.py -i 1 -f 60

   • Use the -i flag with 1 to enable image enhancement using SRCNN. (if you don't want to use it 0 )
   • Adjust the frame interval using the -f flag as needed (default is 60).

3. Once frames are extracted and enhanced (if selected), run the script stitching_image_2_pano.py to stitch the frames into a panorama:

   python stitching_image_2_pano.py

4. The stitched panorama will be saved as stitched_result.jpg in the project directory.

Credits

This project utilizes the SRCNN model for image enhancement. The SRCNN model architecture and weights are based on the work by Dong et al. [1].

References

[1] Dong, C., Loy, C. C., He, K., & Tang, X. (2016). Image super-resolution using deep convolutional networks. IEEE transactions on pattern analysis and machine intelligence, 38(2), 295-307.

Additional Information

Note: The resulting image stitched_result_rooftop.jpg was captured with a Canon mirrorless camera in 4K resolution and stitched using SupreRes-Pano360. The copyright for the photograph belongs to the author.