Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/jvillemare/youdee-finder
Finding the UofD Mascot YoUDee using a TensorFlow ML model.
https://github.com/jvillemare/youdee-finder
python tensorflow tensorflow-models
Last synced: 17 days ago
JSON representation
Finding the UofD Mascot YoUDee using a TensorFlow ML model.
- Host: GitHub
- URL: https://github.com/jvillemare/youdee-finder
- Owner: jvillemare
- Created: 2020-04-06T15:58:27.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2020-04-09T02:36:47.000Z (over 4 years ago)
- Last Synced: 2024-10-11T08:43:48.883Z (about 1 month ago)
- Topics: python, tensorflow, tensorflow-models
- Language: Python
- Homepage: https://villemarette.me/youdee-finder/
- Size: 3.77 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# YoUDee Finder
Locating the University of Delaware mascot, [YoUDee](https://en.wikipedia.org/wiki/YoUDee), is a matter of grave importance.
Bird watching allows us to gain insights into the ecological health of our surroundings. The Delaware Blue Hen is a rare species only found at the University of Delaware.
To aid Delawarean ornithologists, this Machine Learning (ML) model was developed.
[You can also test out the model with your webcam using this webpage here.](https://villemarette.me/youdee-finder/)
## Statistics
The ML model was trained using:
- `1,165` samples of YoUDee images.
- `2,470` samples of neutral, non YoUDee images.
- Neutral samples include other mascots, images of places where YoUDee would be but isn't (football, basketball games), and fursuiters (for their similarity to costumed mascots.)
- `4.37 gigabytes` of image data.
- `632` of the samples were scrapped from YoUDee and other mascot Instagram accounts using [rarcega's scripts](https://github.com/rarcega/instagram-scraper).
Some examples of the samples can be [seen here](samples.md). The full sample data set cannot be provided here due to privacy and copyright concerns.
## Installation
Ensure you are running at least Python 3.7 by typing:
```
python --version
```
Clone this GitHub repo:
```
git clone https://github.com/jvillemare/youdee-finder.git
```
Install the requirements:
`pip install requirements.txt` or `python -m pip install requirements.txt`
## Usage
[Besides using the script locally, you can also use the web-based model here.](https://villemarette.me/youdee-finder/)
Below are the few basic command line flags you can set for this script.
#### Find in a single image
```
python find_youdee.py --image=IMG_0001.jpg
```
#### Find in a single video
```
python find_youdee.py --video=IMG_0001.mov
```
#### Find in a directory
```
python find_youdee.py --dir=pictures/
```
Non-recursively searches a directory for all supported image and video files, then processes and makes all the predictions for them.
#### Write output to a CSV
```
python find_youdee.py --dir=pictures/ --output=predictions.csv
```
Works for both single images and videos, or a full directory.
CSV is in the format of:
- `type`: `image` or `video`.
- `filename`: Filename of image/video.
- `frame_count`: Number of frames in file (images always have `1`).
- `youdee_result`: Confidence that YoUDee is in image as a decimal percent (i.e. `0.50` = 50%).
- `neutral_result`: Confidence that YoUDee is not in an image as a decimal percent.
- `has_youdee`: `True` or `False` boolean, whether or not the confidence of YoUDee was high enough to be significant.
## Other Notes
For videos, YoUDee could be in part of a video, but not other parts. Since we're concerned with finding any instance of YoUDee, if the confidence of YoUDee being in a video is high enough for a consecutive number of frames, then YoUDee is considered to be in that video.