Heat_atttention.py

from PIL import Image
import os 
from clip import CLIP
import pandas as pd
import numpy as np
from utils.utils import (cvtColor, get_configs, letterbox_image,
                         preprocess_input)
                         
def cosine_similarity(vector1, vector2):
    vector1 = np.squeeze(vector1)
    vector2 = np.squeeze(vector2)
    dot_product = np.dot(vector1, vector2)
    norm1 = np.linalg.norm(vector1)
    norm2 = np.linalg.norm(vector2)
    similarity = dot_product / (norm1 * norm2)
    similarity_normalized = 0.5 * (similarity + 1)
    return similarity_normalized


def calculate_single_similarity_score(image_feature, text_feature, scale_factor=np.log(1 / 0.07)):
    # 
    logit_scale = torch.nn.Parameter(torch.ones([]) * scale_factor)
    logit_scale_value = logit_scale.exp()

    # 
    similarity_score = logit_scale_value * (image_feature @ text_feature.t())

    return similarity_score


import json

import numpy as np
import torch
from torch.utils.data import DataLoader
from tqdm import tqdm

from clip import CLIP
from utils.dataloader import ClipDataset, dataset_collate
from utils.metrics import itm_eval


if __name__ == "__main__":
    S_name_ls = ['TCGA-BH-A0B8-01A-02-BSB.64413cea-7a07-43c5-a8d0-939affc290cb.svs', 'TCGA-D8-A1JL-01A-01-TS1.9f253479-0b59-42c5-a7d9-00724cd03439.svs', 'TCGA-BH-A18T-01A-01-TSA.089d9985-6794-4237-a7c7-75c1fb3c1755.svs']
    for S_name in S_name_ls:
        slide_name = S_name
      
        im_name = []
        score_ls = []
        #------------------------------------------------------#
        #------------------------------------------------------#
        datasets_path               = "/heat_map_images/"
        datasets_val_json_path      = "./heat_map_images/{}.json".format(slide_name)
        SVS_name = datasets_val_json_path.split('/')[-1][:-5]
        print(SVS_name)
        
        
        batch_size                  = 1
        num_workers                 = 4
    
        # 
        model       = CLIP()
    
        # 
        val_lines   = json.load(open(datasets_val_json_path, mode = 'r', encoding = 'utf-8'))
        num_val     = len(val_lines)
        # 
        val_dataset = ClipDataset([model.config['input_resolution'], model.config['input_resolution']], val_lines, datasets_path, random = False)
        gen_val     = DataLoader(val_dataset, shuffle=False, batch_size=batch_size, num_workers=num_workers, pin_memory=True,
                                drop_last=False, collate_fn=dataset_collate, sampler=None)
    
        # 
        i_features = []
        t_features = []
        with open(datasets_val_json_path, 'r') as f:
            data_ls = json.load(f)
        X = 0
        for iteration, batch in tqdm(enumerate(gen_val)):
            images, texts = batch
            file_name = data_ls[X]['image'].split('/')[-1]
            with torch.no_grad():
                if model.cuda:
                    images  = images.cuda()
                images_feature, texts_feature = model.detect_image_for_eval(images, texts)
                images_feature = images_feature #.cpu().numpy()
                texts_feature = texts_feature   #.cpu().numpy()
                
                
                
                
                # score = cosine_similarity(images_feature,texts_feature)
                # -------------------
                score_similiraty = calculate_single_similarity_score(images_feature,texts_feature)
                score = score_similiraty.cpu().item()  # 1 / (1 + torch.exp(-score_similiraty)).cpu().item()
                # -----------------
                
                print('file_name',file_name,"score", score)
                im_name.append(file_name)
                score_ls.append(score)
                X = X+1
        dict_data={'name':im_name,'score':score_ls}
        df = pd.DataFrame(dict_data)
        csv_file = "./heat_map_images/model_score_{}.csv".format(SVS_name)
        df.to_csv(csv_file, index=False)
        print(csv_file,'-----Finished')