2320sharon / sniffer Goto Github PK

View Code? Open in Web Editor NEW

21.0 1.0 4.0 16.64 MB

Lightweight Python application for sorting images in your dataset

License: MIT License

Jupyter Notebook 54.39% Python 45.61%

python panel holoviz image-classification

sniffer's Introduction

Sniffer 🐕

"Sniff out the good and bad imagery in your dataset."

A python application for sorting through imagery.
Only sorts jpgs in the images directory
Saves the sorted jpgs in two directories called good_images and bad_images within the program's directory when File Mode is activated
Saves the outputs as a csv into the programs current working directory when CSV Mode is activated

Running the Program 💻

There are two ways to run the program, from the command line or from within a jupyter notebook.

Jupyter Notebook Method 🪐

conda activate Sniffer
cd <location where you saved Sniffer>
jupyter notebook

Command Prompt Method

conda activate Sniffer
cd <location where you saved Sniffer>
panel serve Sniffer.ipynb
Copy the local host text into your browser of choice http://localhost:5006/Sniffer

How to Use Hotkeys ⌨️

I recommend using hotkeys in the localhost version of Sniffer. Sometimes jupyter notebooks will create and delete cells when you type a and d. That being said hotkeys will still work in jupyter notebook.

Sniffer's Hotkeys

a = good image
d = bad image
s = undo last action

Installation Guide & Updating 🧰

Download Sniffer ⏬

Clone the repository

git clone --depth 1 https://github.com/2320sharon/Sniffer.git

(--depth 1 means "give me only the present code, not the whole history of git commits" - this saves disk space, and time)

Install Sniffer 🧰

conda env create --file install/sniffer.yml
conda activate Sniffer

Update Sniffer ⏫

cd <location where you installed Sniffer>
cd install
conda activate sniffer
conda env update --file sniffer.yml --prune

Directory Structure

├── Sniffer
│   ├── Sniffer
│   │   ├── tests                       #Don't touch the test directory
|   |   |  |_ test_data
|   |   |_ __init__.py
|   |   |_ conftest.py
|   |   |_ test_sniffer.py
|   | 
│   ├── __init__.py
│   ├── sniffer.py
│   
├── install
|    |_sniffer.yml
├──thumbnails
|    |_ <OTHER THUMBNAILS GENERATED BY SNIFFER>
|
|___images
|    |_ <YOUR IMAGES HERE>
|
|___good_images
|    |_ <GOOD IMAGES MOVED HERE BY SNIFFER>
|
|___BAD_images
|    |_ <BAD IMAGES MOVED HERE BY SNIFFER>
|
├── README.md
├── assests
├── .github
└── .gitignore

Quick Disclaimer

Currently sniffer only works with images with the .jpg extension. Other file types are not supported.

Unsupported File Types

❌ .PNG

😮 Updates Coming Soon 😮

Support for .pngs
Hosting 🌐
A wiki full of examples and troubleshooting tips

sniffer's People

Stargazers

Watchers

Forkers

dnschavarria stjordanis mardascience cameronbodine

sniffer's Issues

Image classification workflow dataset

Here's a small dataset to play with image classification

Sniffer_Output_17_03_2022_hr_17_03.csv

images

and this looks like a really nice potential method, or at least a full workflow to get you started. you would have to modify the way the images were read in from csv, but otherwise it could be similar (simpler, because only 2 classes "good" and "bad") 0 we could chat further

Non-square imagery and Panel App

non square imagery did not render at the correct aspect ratio on the panel app (Linux + Firefox).
It did work w/ correct aspect ratio on the Jupyter Notebook (but its harder to make the image larger on jupyter notebook b/c of margins of the notebook)

example image is provided, as well as screenshot of panel app and notebook

example image:

Panel App:

Notebook:

And for completeness.. streamlit sniffer preserves the correct aspect ratio:

Beyond 'good' and 'bad' (enhancement)

How hard would it be to replace the yes/no buttons with other options (and perhaps beyond 2).. I am thinking of a Likert scale (1-5)?
I could see a few ways to do this:

a free text field, where people put a number?
a slider where people could rate images between 0 and 1?
multiple buttons (5 in this case, vs 2 in yes/no)?
radio buttons (which i guess is like the multiple button idea)?

Sniffer Doesn't Accept `JPEG`s

Sniffer will convert .JPG to .jpg, but not .jpegs.
There needs to be a generic conversion function that converts any jpg type of image to the .jpg. The same function could be used for .pngs as well.

REGEX would be great to use to scan the file extensions for matches

Create Automated Testing Suite

Update ReadMe

The ReadMe doesn't have enough details about how to run and install Sniffer.

Add more detailed installation instructions
Mention that currently it only runs in localhost or in jupyter notebook
How to create the anaconda environment
Update the recording
Update the roadmap

Upgrade Sniffer's Documentation

Upgrade Sniffers Documentation

Display "number of images left to tag"

I like how there is a text display that increments to tell me how many images I have tagged. But I'd also like to know how many images I have left to do. The principal reason is because eventually I'd like there to be an optional button that would initiate model training to autocomplete the classification process. Without a "X images remaining" display, a user wouldn't easily gauge when to stop tagging and start model training and implementation

this is like "human in the loop" for binary (yes/no type) image classification.

Sniffer Doesn't Accept PNGs

Currently, Sniffer can only render .jpg files. Sniffer needs to be able to detect .png files and render a png panel if it finds them.

Sniffer needs to detect .png files and render a png panel
Sniffer needs to hide the jpg panel and show the png panel instead when a .png file is found
Sniffer needs to convert file types .PNG to .png