555

[dreamlychina]

import torch
import clip
from PIL import Image
from torchvision import transforms
import json
import cv2
import numpy as np
from sklearn.cluster import OPTICS, KMeans
from sklearn.manifold import TSNE
from sklearn.datasets import make_blobs
import matplotlib.pyplot as plt
import matplotlib
import random
matplotlib.use('TkAgg')

import os
import pdb

all_img_text_features = {"coco/train2017/":[], "vg/VG_100K/":[], "vg/VG_100K_2/":[],
"gqa/images/":[], "ocr_vqa/images/":[], "textvqa/train_images/":[]}

all_img_path = {"coco/train2017/":[], "vg/VG_100K/":[], "vg/VG_100K_2/":[],
"gqa/images/":[], "ocr_vqa/images/":[], "textvqa/train_images/":[]}

if name == "main":
device = "cuda" if torch.cuda.is_available() else "cpu"
model, processor = clip.load("ViT-B/32", device=device)
tokenizer = clip.tokenize

input_json = "llava_v1_5_mix665k.json"
# 从文件解析JSON
merged_datas = []
count = 0
batch_size = 2

img_paths = []
all_descriptions = []

with open(input_json, 'r', encoding='utf-8') as json_file:
    datas = json.load(json_file)
    print("all data num: ", len(datas))
    #datas = random.sample(datas, 103)
    print("all data num: ", len(datas))
    max_num = -1
    max_num_list = []
    max_num_img_path = []
    for data in datas:
        img_text_id = data['id']

        if "image" not in data.keys():
            count +=1
            img_path = os.path.join("textvqa/train_images", img_text_id+".jpg")
        else:
            img_path  = data['image']
        print(img_path)

        #test
        #img_path = "med_1.png"
        #img = Image.open(img_path)
        conversations = data['conversations']
        descriptions = ""
        for index, conversation in enumerate(conversations):
            #print("num: ", len(conversation['value'].split(" ")))
            if len(conversation['value'].split(" ")) > max_num:
                max_num = len(conversation['value'].split(" "))
                max_num_list.append(max_num)
                max_num_img_path.append(img_path)
    print("max_num: ", max_num_list)
    print("max_num_img_path: ", max_num_img_path)