fabfile.py

# Libraries
import os

import yaml
from fabric import task

from src.keypoints import KeypointsDetector
from src.keypoints import SignPredictor
from src.keypoints.mpLoader import MediaPipeLoader
from src.preprocess import CutFrames, SampleAugment, Keypoints2Dataset

from src.train import DatasetLoader, Trainer, Tester

# Read tasks and paths from config file
with open('config.yaml', 'r') as f:
    config = yaml.safe_load(f)

#-------------------------------------------------------------------

@task
def CutLastFrame(c):
    problematic_videos = [
        "data/CrossDatasetVideos/melancia/1/signbank_MELANCIA.mp4",
        "data/CrossDatasetVideos/misturar/1/signbank_MISTURAR.mp4",
        "data/CrossDatasetVideos/nadar/1/signbank_NADAR.mp4",
        "data/CrossDatasetVideos/patins/1/signbank_PATINS.mp4"
    ]
    CutFrames.cut_problematic_frame(problematic_videos)

@task
def ProcessVideos(c):
    dataset_videos_path = config['tasks']['ProcessVideos']['dataset_videos_path'] 
    dataset_frames_path = config['tasks']['ProcessVideos']['dataset_frames_path']  
    number_of_frames    = 15          # Number of frames to sample from each video
    size                = (640, 480)  # Size to resize the frames
    augment_factor      = 4          # Number of augmented frames per original frame
    random_state        = 77782983
    show                = False
    SampleAugment.process_videos(dataset_videos_path,dataset_frames_path,number_of_frames,size,augment_factor,random_state,show)
     
@task
def ExtractKeypoints(c):
    dataset_frames_path = config['tasks']['ExtractKeypoints']['dataset_frames_path']
    dataset_path        = config['tasks']['ExtractKeypoints']['dataset_path'] 
    modes               = ["coordinates", "angles_distances"]
    media_pipe_loader   = MediaPipeLoader() 
    Keypoints2Dataset.process_frames(dataset_frames_path, dataset_path, modes[1], media_pipe_loader, show=False)

@task 
def SplitDataset(c):
    dataset_path        = config['tasks']['SplitDataset']['dataset_path']
    series_size         = 15  # Number of frames Samples
    train_percentage    = 0.8
    val_percentage      = 0.19999
    test_percentage     = 0.00001
    random_state        = 77796983
    DatasetLoader.split_dataset(dataset_path, train_percentage, val_percentage, test_percentage, series_size, random_state)

@task 
def kFold(c):
    dataset_path        = config['tasks']['SplitDataset']['dataset_path']
    series_size         = 30  # Number of frames Samples
    train_percentage    = 0.8
    val_percentage      = 0.2
    random_state        = 77796983
    DatasetLoader.split_dataset_kfold(dataset_path, train_percentage, val_percentage, series_size, random_state)
    
#-------------------------------------------------------------------

@task
def TrainModel(c):
    train_dataset_path   = config['tasks']['TrainModel']['train_dataset_path']
    val_dataset_path     = config['tasks']['TrainModel']['val_dataset_path']
    output_folder_path   = config['tasks']['TrainModel']['output_folder_path']
    series_size          = 15
    Trainer.train(train_dataset_path, val_dataset_path, output_folder_path, series_size)

@task
def TestModel(c):
    test_dataset_path  = config['tasks']['TestModel']['test_dataset_path']
    model_path         = config['tasks']['TestModel']['model_path'] 
    series_size        = 15
    Tester.test(test_dataset_path, model_path, series_size)

#-------------------------------------------------------------------

@task
def LiveHandTrack(c):
    window_size        = (1080, 720)
    media_pipe_loader  = MediaPipeLoader()
    KeypointsDetector.live_hands_tracking(window_size, media_pipe_loader) 
    
@task
def LiveSignDetect(c):
    window_size        = (1080, 720) # Size to render image
    frame_size         = (640, 480)  # Size to resize the frames
    threshold          = 0.5         # Minimum confidence to predict sign
    media_pipe_loader  = MediaPipeLoader()
    model_path         = config['tasks']['TestModel']['model_path']
    sign_predictor = SignPredictor.SignPredictor(model_path)
    SignPredictor.live_sign_detection(window_size, frame_size, threshold, media_pipe_loader, sign_predictor)

@task
def VideoSignDetect(c):
    frame_size         = (640, 480)  # Size to resize the frames
    media_pipe_loader  = MediaPipeLoader()
    model_path         = config['tasks']['TestModel']['model_path']
    sign_predictor = SignPredictor.SignPredictor(model_path)
    pred = sign_predictor.predict_from_video("data/Esquecer.mp4", frame_size, media_pipe_loader)
    print(sign_predictor.get_top_i_predictions(pred, 5))

from src.utils import utils_files
@task
def create(c):
    path = "data/OurDatasetFilesCSV_15/angles_distances/full_dataset.csv"
    # utils_files.remove_first_row_from_csv(path)
    utils_files.add_row_to_csv(path, [49, 89])