The aim is to stitch multiple images together and create a panorama of stitched images.

Task

Compare two input images and return a value that tells how visually similar the given images are. The lower the score, the more contextually similar the two images are with a score of '0' being identical.

Task:

• Create a custom logo for deeppixel taking into consideration the themes of deep learning, open source and tech
  • Name it as deeppixel_icon.jpeg
  • Please provide a high-resolution image as well as an icon
  • change your branch to gh-pages and put it inside the assets folder
  • Give a Pull Request to this repo's gh-pages branch
• Create a template image for the GitHub repo in accordance to the above theme
  • Image size should be 1280×640px
  • Name it as deeppixel_template.jpeg
  • change your branch to gh-pages and put it inside the assets folder
  • Give a Pull Request to this repo's gh-pages branch

The script should detect the text in images automatically

• You may or may not use Deep Learning for this

Suggested Tasks:

• Cartoonify
• Oldify
• Black and White to Color
• Pencil Sketch - Black and White
• Pencil Sketch - Color
• Oil Painting
• Water Color
• Pointilism

References:

• ArtCV
• White Box Cartoonization: GitHub, YoutTube Explanation, Paper

** For now, we prefer to use TensorFlow [ Later on we will include support for PyTorch as well]

Task:

Blur the faces in an image as well as video

Suggested workflow:

• Inside the deeppixel directory, create a new sub-directory face_blur[Please name it appropriately and use camel_case]
• In the first attempt 💭💭 use a [Jupyter notebook] to perform your work.
• Once you are done give a Pull Request🩹 with the message 📩Developed Jupyter Notebook for Face Blur , briefing about your approach in the description and add a link of the above notebook in Google Colab [Please ensure you have given access] ⛔
• Once merged😎 , build a script for the same task in the face_blur directory __(If you are using Deep Learning, ensure that you have saved your trained model and its weights so that in the script you build can simply fetch it instead of training again)
• Use argparse library so that the input image and the output path can be given as arguments in the terminal while running the script
• Update the requirements.txt file in the root directory of the master branch to ensure any additional modules you have used in present there.
• Make sure you provide sample images/videos 📷 used
• Give a Pull Request 🩹 with the message 📩Developed Script for Face Blur and mention how you have given the argument parameters to run the script in the description
• Once approved, work on documenting every block of code if not every line of your script
• Add a README.MD file with appropriate description [Please ensure you properly cite any research paper or blog you have taken direct reference from]
• Give another Pull Request 🩹 with a message 📩 : Documentation Updated for Face Blur
• Now save the model and the model weights, build a single python script that takes in an image and gives us the output (Make sure the model and model weight is properly named for future use) by using your already trained model

References:

• PyImageSearch blog

Additional Tasks:

• Curate a custom dataset 🧰 for this task
• Look for better methods to improve 🥇 this

Task

Take input a content image,style image and ]output a image that transfers the style into the content image

Guidelines:

• Experiment building this using a Jupyter notebook locally or on Google Colab
• Once its giving desired results put it inside the artistic_style_transfer directory
• Build a script for the task inside the bw_to_c directory
• Inside the artistic_style_transfer directory, create two folders input and output to be used for the input and output images
• Use argparse library so that the input image and output path can be given as arguments in the terminal while running the script
• Create a requirements.txt file and specify the modules used
• Try your script/notebook with multiple images and store the results in output folder
• Comment your script/notebook well
• Create/Update the ReadME.MD file:
  • Name of the task
  • A small description
  • Approach used
  • Input and Output Images
• Make sure you have the model weight or any related files to run the script in the same directory
• Push the changes and give a Pull Request
• In your PR, please try to give a link to a Colab Notebook(if applicable) as a comment.
• In your PR, put a reference to the issue it is for
• once your model/script works, curate your own dataset and get the outputs..(Try 10 images) and give a second PR about your own Dataset
• Try out other methods to implement the above

Please do not use a code from someone else's repo or a blog like PyImageSearch directly..You can definitely refer to others' code. But make sure you make some contributions of your own into it. If you strongly use a code from someone else,please credit them properly in the README file.

Task:

Take in a low very low-resolution image and output an enlarged and high resolution image

Suggested workflow:

• Inside the deeppixel directory, create a new sub-directory img_super_res[Please name it appropriately and use camel_case]
• In the first attempt 💭💭 use a [Jupyter notebook] to perform your work.
• Once you are done give a Pull Request🩹 with the message 📩Developed Jupyter Notebook for Image Super-Resolution, briefing about your approach in the description and add a link of the above notebook in Google Colab [Please ensure you have given access] ⛔
• Once merged😎 , build a script for the same task in the img_super_res directory __(If you are using Deep Learning, ensure that you have saved your trained model and its weights so that in the script you build can simply fetch it instead of training again)
• Use argparse library so that the input image and the output path can be given as arguments in the terminal while running the script
• Update the requirements.txt file in the root directory of the master branch to ensure any additional modules you have used in present there.
• Make sure you provide sample images/videos 📷 used
• Give a Pull Request 🩹 with the message 📩Developed Script for Image Super-Resolution and mention how you have given the argument parameters to run the script in the description
• Once approved, work on documenting every block of code if not every line of your script
• Add a README.MD file with appropriate description [Please ensure you properly cite any research paper or blog you have taken direct reference from]
• Give another Pull Request 🩹 with a message 📩 : Documentation Updated for Image Super-Resolution

References :

Will be updated with more soon, Meanwhile, feel free to try these

• ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks

  • Paper
  • Available Implementation in TensorFlow
  • Available Implementation in PyTorch

• Second-Order Attention Network for Single Image Super-Resolution

  • Paper
  • Available Implementation

• Image Reconstruction with Predictive Filter Flow

  • Paper
  • Available Implementation

Additional Tasks:

• Curate a custom dataset 🧰 for this task
• Look for better methods to improve 🥇 this

• Brisque
• RankIQA
• DeepIQA

Task:

Create custom GitHub Actions or use pre-existing ones to automate the workflow for this project

Analyse the images taken from the webcam and detect any movement if present.

Task:

Enhance images taken in low light conditions

Suggested workflow:

• Inside the deeppixel directory, create a new sub-directory img_matting[Please name it appropriately and use camel_case]
• In the first attempt 💭💭 use a [Jupyter notebook] to perform your work.
• Once you are done give a Pull Request🩹 with the message 📩Developed Jupyter Notebook for Image Matting , briefing about your approach in the description and add a link of the above notebook in Google Colab [Please ensure you have given access] ⛔
• Once merged😎 , build a script for the same task in the img_matting directory __(If you are using Deep Learning, ensure that you have saved your trained model and its weights so that in the script you build can simply fetch it instead of training again)
• Use argparse library so that the input image and the output path can be given as arguments in the terminal while running the script
• Update the requirements.txt file in the root directory of the master branch to ensure any additional modules you have used in present there.
• Make sure you provide sample images/videos 📷 used
• Give a Pull Request 🩹 with the message 📩Developed Script for Image Matting and mention how you have given the argument parameters to run the script in the description
• Once approved, work on documenting every block of code if not every line of your script
• Add a README.MD file with appropriate description [Please ensure you properly cite any research paper or blog you have taken direct reference from]
• Give another Pull Request 🩹 with a message 📩 : Documentation Updated for Image Matting
• Now save the model and the model weights, build a single python script that takes in an image and gives us the output (Make sure the model and model weight is properly named for future use) by using your already trained model

References :

• The best method so far per

• The official implementation

This issue is related to a fairly complex topic and hence open to everyone for the contribution!

• Initially, try with images and then proceed with videos if possible

Create class activation maps to explain the black-box nature of convolutional neural networks. Remember, that this is primarily aimed at customed trained models rather than pre-trained models so that we can input a h5 or saved_model and an image and analyze.

Tasks ( Implement the following):

• GradCAM
• SmoothGrad
• GradCAM++
• ScoreCAM

Suggestions:

• Please cite all references
• The code should be properly documented for each and every section
• Modularize into functions as much as possible
• Use meaningful docstrings
• Build it as a module cam

Task:

Enhance images taken in low light conditions

Suggested workflow:

• Inside the deeppixel directory, create a new sub-directory img_undark[Please name it appropiately and use camel_case]
• In the first attempt 💭💭 use a [Jupyter notebook] to perform your work.
• Once you are done give a Pull Request🩹 with the message 📩Developed Jupyter Notebook for Enhancement of Low Light Images , briefing about your approach in the description and add a link of the above notebook in Google Colab [Please ensure you have given access] ⛔
• Once merged😎 , build a script for the same task in the img_undark directory __(If you are using Deep Learning, ensure that you have saved your trained model and its weights so that in the script you build can simply fetch it instead of training again)
• Use argparse library so that the input image and the output path can be given as arguments in the terminal while running the script
• Update the requirements.txt file in the root directory of the master branch to ensure any additional modules you have used in present there.
• Make sure you provide sample images/videos 📷 used
• Give a Pull Request 🩹 with the message 📩Developed Script for Enhancement of Low Light Images and mention how you have given the argument parameters to run the script in the description
• Once approved, work on documenting every block of code if not every line of your script
• Add a README.MD file with appropriate description [Please ensure you properly cite any research paper or blog you have taken direct reference from]
• Give another Pull Request 🩹 with a message 📩 : Documentation Updated for Enhancement of Low Light Images
• Now save the model and the model weights, build a single python script that takes in an image and gives us the output (Make sure the model and model weight is properly named for future use) by using your already trained model

References :

• Learning to see in the dark research paper

• The official implementation

• An explanatory blog

Additional Tasks:

• Curate a custom dataset 🧰 for this task
• Look for better methods to improve 🥇 this

Do it inside the sub-directory:
https://github.com/smaranjitghose/DeepPixel/tree/master/deeppixel/denoise

Try to take in your own images and apply the same.

Make sure you upload the images used in a images folder

A jupyter notebook is preferred but a script works well too

Please comment each part of your code, explaining why it's used so that others can use this as a groundwork for future work

Panoptic segmentation unifies the typically distinct tasks of semantic segmentation (assign a class label to each pixel) and instance segmentation (detect and segment each object instance).

To get started,
Refer this link

• Photorealistic Style Transfer
• Artistic Style Transfer

Tasks:

• Work on the readme and make it more interactive and add on issues and other things from time to time
• Make sure the READMEs of the individual modules are synchronous
• Work on the Github Page for this project gh-pages branch
• Learn and use Sphinx for actual documentation of this project as a package
• Make a template for Giving Pull Requests