generate_predictions.py

#%%
import sys
import os
# sys.path.insert(0, os.path.abspath("C:\\Users\Anirbit\\L4 Project\\L4-Project\\src"))
sys.path.insert(0, os.path.abspath("/home/anirbit/Anirbit_Ghosh/L4-Project/src"))
from data_loading.data_transform import validation_transfomer, pretrained_pred_transformer
from models.custom_network import Net
import torch
import matplotlib.pyplot as plt
from torchvision import models
import torch
import torch.nn as nn
from PIL import Image
import pandas as pd
from pathlib import Path
import numpy as np
import pandas as pd
import argparse

#%%
def dir_path(string):
    if os.path.isdir(string):
        return string
    else:
        raise NotADirectoryError(string)
    
def file_path(string):
    if os.path.isfile(string):
        return string
    else:
        raise FileNotFoundError(string)
    
#%%
parser = argparse.ArgumentParser()
parser.add_argument('-n', '--net', type=file_path, help='Pass path to file containing model weights with -i or --image flags' )
parser.add_argument('-t', '--tiles', type=dir_path, help="Pass directory containing tiles to predict lables of using -l or --label flags")
parser.add_argument('-o', '--output', type=dir_path, help="Pass output directory using -o or --output flags")
parser.add_argument('-w', '--whole', type=dir_path, help="Pass whole slide input directory using -w or --whole flags")

device = "cuda:0" if torch.cuda.is_available() else 'cpu'

# %%
def read_model(weights):
    params_model = {
        "input_shape" : (3, 96, 96),
        "initial_filters" : 8,
        "num_fc1" : 100,
        "dropout_rate" : 0.25,
        "num_classes" : 2,
        "activation_func" : 'tanh',
    }
    
    model = Net(params_model)
    checkpoint = torch.load(weights)
    model.load_state_dict(checkpoint)
    
    return model

def get_class_predictions(model, tile_dir, image_dir, output_dir):
    transform  = validation_transfomer()
    
    print("Getting each whole slide image name...")
    for file in os.listdir(image_dir):
        dirname = file[:-4]
        print(f"Processing predictions for image : {dirname}")
        tile_path = os.path.join(tile_dir, dirname)
    
        predictions = []
        print("Starting prediction generation for tiles...")
        for image_file in os.listdir(tile_path):
            image_name = image_file[:-4]
            image = Image.open(os.path.join(tile_path, image_file))
            print("Predicting class for image {} ...".format(image_name))
            input = transform(image)
            
            output = model(input)
            prediction = int(torch.max(output.data, 1)[1].numpy())
            print("Prediction success - saving output!")
            predictions.append(prediction)

        df = pd.DataFrame({"image" : os.listdir(tile_path), "predictions" : predictions})
        csv_name = dirname + "-predictions.csv"
        df.to_csv(os.path.join(output_dir, csv_name))     


#%%
def get_probability_predictions(model, tile_dir, image_dir, output_dir):
    transform  = validation_transfomer()
    print("Getting each whole slide image name...")
    for file in os.listdir(image_dir):
        dirname = file[:-4]
        print(f"Processing predictions for image : {dirname}")
        tile_path = os.path.join(tile_dir, dirname)
    
        predictions = []
        print("Starting prediction generation for tiles...")
        for image_file in os.listdir(tile_path):
            image_name = image_file[:-4]
            image = Image.open(os.path.join(tile_path, image_file))
            print("Predicting class for image {} ...".format(image_name))
            input = transform(image)
            
            output = model(input)
            prediction = np.exp(float(output.data[0][1].numpy()))
            print("Prediction success - saving output!")
            predictions.append(prediction)

        df = pd.DataFrame({"image" : os.listdir(tile_path), "predictions" : predictions})
        csv_name = dirname + "-predictions.csv"
        df.to_csv(os.path.join(output_dir, csv_name))     
  
#%%
if __name__ == "__main__" :
    if len(sys.argv) < 6:
        parser.print_usage()
        parser.print_help()
        sys.exit(1)
        
    print("Parsing arguments...")
    args = parser.parse_args()
    print("Parsing arguments... DONE")
    
    print("Locating directories...")
    net_path = args.net
    tile_dir = args.tiles
    out_dir = args.output
    
    image_dir = args.whole
    print("Locating directories... DONE!")
    
    print("Generating model with given weights... ")
    model = read_model(net_path)
    print("Generating model with given weights... DONE!")
    
    # get_class_predictions(model=model, tile_dir=tile_dir, image_dir=image_dir, output_dir=out_dir)
    get_probability_predictions(model=model, tile_dir=tile_dir, image_dir=image_dir, output_dir=out_dir)