[G05] Computer Vision Project 2020

To run use the command python proj.py

Options:

• [Required] -i or --input followed by the path of the input file
• -o or --output followed by the path of the desired output file without extension. If it is not defined the result will be shown in a window.
• -v or --verbose, if defined will run the project in verbose mode
• -s or --skip followed by a number defines the number to frame to skip - 1. For example if value is 3 it will analyze 1 frame each 3.
• -d or --debug will enable debug-mode

Setup

The code is found under the code folder.

Required python modules are found in code/requirements.txt and therefore can be installed with pip install -r requirements.txt

The only file that we were not able to re-create or retrieve during runtime are YOLO network weights, that must be downloaded here(our drive) or here(developer site) and put into code/detection/people/yolo-coco/.

Output

The output of the project consists in:

1. An output video in which we annotate:
   • For each painting bounding box, the ROI containing the rectified version of the original painting and the indication of the first retrieved db painting
   • For each person, the ROI with the indication of the relative room
2. A terminal output in which we print:
   • The description of what the program is processing (ONLY IF VERBOSE MODE IS ACTIVATED)
   • For each frame, and for each painting detected, the indication of the corresponding bounding box and the entire result of the ranked list retrieved (ALWAYS)
   • For each frame, and for each person detected, the indication of the corresponding bounding box (ALWAYS)
   • For each frame the detected room (ALWAYS)

Paper

Detail of the pipline can be found inside the paper document

Example videos

These will open youtube videos

Example video #1

Example video #2

Example video #3

Example video #4

Example video #5