Graduation Project Code - Changelog

Table of Contents

• Week 11
• Week 10
• Week 9
• Week 8
• Week 7
• Week 5
• Week 4
• Week 3
• Week 2

Week 11

Mar 25:

• Restructured Saved Weights Folder
• Added 6 Models trained on data of 13 words and 4 people
  • 5 FPS x2 (Max Accuracy and Min Loss)
  • 9 FPS x2 (Max Accuracy and Min Loss)
  • 13 FPS x2 (Max Accuracy and Min Loss)
• Added Notebooks that were used for training and recording data

Mar 21:

• Added 4 Models trained on data of 13 words and 3 people

  • 25 FPS x2 (Max Accuracy and Min Loss)
  • 13 FPS
  • 9 FPS

• Added 2 Models trained on data of 13 words and 4 people

  • 25 FPS
  • 13 FPS

Week 10

Mar 19:

• Alphabets added to NLP Dataset
• Added new NLP Model and Weights trained on new dataset

Mar 15:

• Added NLP weights to handle "Hello [Name]"

Mar 14:

• Added 2 Models trained on data of 10 words and 4 people
• Updated Script to Record CV Data - CV_Model.ipynb

Week 9

Mar 6:

• Dataset changes
  • Combined 2 datasets (Dataset 1 and Dataset 2) into one dataset - Dataset_1780_rows.csv
  • Created a double version of the dataset
  • Fixed finger spelling placeholders
  • Added new sentences
• Updated NLP Script
  • Handles finger spelling now
  • Adds '?' after 6 W's and How
  • Fixed hardcoded values when initialing model
• Updated weights and myVars.txt for the NLP model

Week 8

Mar 5:

• Updated NLP Script
  • Added function preprocess_sentence that will handle all preprocessing
  • Added code to handle finger spelling
  • Handling '?' (QM-wig) and manually adding ? to decoded sentence

Mar 3:

• Computer Vision Data Collection script updated
  • Added code to prompt action on screen while collecting data
• Uploaded the Updated version of Main Notebook - Updated.ipynb
• Uploaded new models in Saved Weights folder

Mar 1:

• Computer Vision data collection script changes
  • Removed useless Imports
  • Changed draw_landmarks (Face landmarks)
  • Added function (extract_keypoints)
  • Changed sequence length to 25 and number of sequences to 40

Feb 28:

• Added 2 new datasets for NLP

Week 7

Feb 22:

• Uploaded 4 new models with training Camera position as Standing and Seated
• Updated NLP_Model.py so that it handles words that aren't part of the dictionary

Feb 21:

• Created new issues and assigned who needs to work on them
• Removed all additional Datasets

Week 5

Feb 12:

• Added the Computer Vision Model folder

  • With Notebooks folder containing the main notebook and the saved weights
  • With Scripts folder containing 3 scripts for training data collection

• Restructured the NLP Model folder

  • Reduced the folder hierarchy for better easy navigation
  • Added a ReadMe file

Feb 11:

• Restructured the repository
• Added 2 Main folders for each Model; NLP Model and Computer Vision Model.

Feb 10:

• Added ReadMe file and small update to NLP_Model.py
  • ReadMe file for the Scripts/Model directory that shows how to use the NLP Model
  • NLP_Model.py now has a loop that keeps taking input

Feb 8:

• Major update to NLP_Model.py
  • Combined generate_batch function with decode_sequence and now it takes an user input instead of a dataset
  • Made it so that the python program can take an argument, this will be the ASL text
  • Removed unused imports
  • Added new colors for printing text
  • Removed code for importing and preprocessing dataset
  • Removing import to other file with functions, only 1 function was being used so it has now been moved to the file itself (read_list_from_file)
  • Added timing for all sections of the code
  • Removed all unnecessary code

Feb 6:

• Added a script for NLP model
  • Reorganized Scripts folder into 2 parts; Data & Model
    • Scripts/Data holds all the scripts related to data extraction
    • Scripts/Model holds the python code for the NLP model
  • Updated Encoder_Decoder_V4 file to save model in H4 format
  • Added Scripts/Model folder
    • NLP_Model.py: Refactored the python notebook code
    • functions.py: Unimportant functions moved to this
    • myVars.txt: Dependant variables that take the longest time to load are now saved in here (over a pkl file for readability)
    • Incomplete task: Change the decoder function code to take user inputs

Week 4

Feb 5:

• Added Encoder_Decoder V4
  • Removed .pkl files as they are not required (check commit description for more details)
  • Added a file with the dataset rows doubled for testing

Week 3

Jan 25:

• Dataset Addition from this site https://www.lifeprint.com/asl101/index/sign-language-phrases.htm
• Made changes to the Encoder_Decoder Notebook for the PoC Presentation

Jan 23:

• Dataset changes
  • Removing all maths sentences
  • Some sentences had 3 lines (removing those sentences)
• Fix for removingSpacesFromDataset.py: Previous code wasn't iterating properly and was pasting the same sentences twice
• Reorganized repository

Jan 22: Added Encoder_Decoder_V2.ipynb: 2nd Version of EncoderDecoder

Jan 20: New dataset added, has around 2K rows of data & script written by her

Week 2

Jan 19: Updated python notebook with completed encoder-decoder code. Still needs a lot of improvements

Jan 17: Repo set up