with Gemini 2.5 online/InternVL3 offline

Using Gemini 2.5 online

Code

from google import genai
import keyring
from PIL import Image
import torchvision.transforms as T
from torchvision.transforms.functional import InterpolationMode
import sys
import pickle
import math
import numpy as np
import torch

Code

client = genai.Client(api_key=keyring.get_password("system", "google_ai_api_key"))

list model

Code

print("List of models:\n")
for m in client.models.list():
    for action in m.supported_actions:
       # if action == "generateContent":
            print(m.name+" "+ action)

English Extract

Code

image = Image.open("images/english.jpg")

response_gemini_en = client.models.generate_content(
    model="gemini-2.5-pro-exp-03-25",
    contents=[image, "Extract text from image"])

Code

print(response_gemini_en.text)

Write slowly and take the time
to make sure each letter
is the perfect shape

chinese Extract

Code

image = Image.open("images/chinese.png")

response_gemini = client.models.generate_content(
    model="gemini-2.5-pro-exp-03-25",
    contents=[image, "提取图上的文字"])

Code

print(response_gemini.text)

放养

把我不羁的灵魂
放养在可可西里的草原上，
藏雪狐活泼没人爱人
我与它捉迷藏

把我不羁的灵魂，
放养在撒哈拉沙漠上，
看生命在贫瘠的土地上，
依然欣欣向荣地生长

把我不羁的灵魂
放养在昏黄遗迹的小岛
坐上星期五的木筏
勇敢地乘风破浪

把我不羁的灵魂
放养在天涯海角
就让我自由地去流浪。

Using InternVL3 1B model offline

https://huggingface.co/OpenGVLab/InternVL3-8B-hf

If using better model will increase accuracy

Code

print(sys.version)

Code

pip install --upgrade transformers
pip install einops timm
pip install -U bitsandbytes

Code

import os
os.system('pip show transformers')

Code

import torch
from transformers import AutoTokenizer, AutoModel,pipeline
path = "OpenGVLab/InternVL3-1B"

model = AutoModel.from_pretrained(path,torch_dtype=torch.bfloat16,trust_remote_code=True).eval()

tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)

Code

generation_config = dict(max_new_tokens=1024, do_sample=True)

testing

Code

question = 'Hello, who are you?'
response, history = model.chat(tokenizer, None, question, generation_config, history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

single-image single-round conversation (单图单轮对话)

Code

IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([
        T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=MEAN, std=STD)
    ])
    return transform

def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio

def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
        i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images

def load_image(image_file, input_size=448, max_num=12):
    image = Image.open(image_file).convert('RGB')
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values

def split_model(model_name):
    device_map = {}
    world_size = torch.cuda.device_count()
    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
    num_layers = config.llm_config.num_hidden_layers
    # Since the first GPU will be used for ViT, treat it as half a GPU.
    num_layers_per_gpu = math.ceil(num_layers / (world_size - 0.5))
    num_layers_per_gpu = [num_layers_per_gpu] * world_size
    num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.5)
    layer_cnt = 0
    for i, num_layer in enumerate(num_layers_per_gpu):
        for j in range(num_layer):
            device_map[f'language_model.model.layers.{layer_cnt}'] = i
            layer_cnt += 1
    device_map['vision_model'] = 0
    device_map['mlp1'] = 0
    device_map['language_model.model.tok_embeddings'] = 0
    device_map['language_model.model.embed_tokens'] = 0
    device_map['language_model.output'] = 0
    device_map['language_model.model.norm'] = 0
    device_map['language_model.model.rotary_emb'] = 0
    device_map['language_model.lm_head'] = 0
    device_map[f'language_model.model.layers.{num_layers - 1}'] = 0

    return device_map

English Extract

Code

pixel_values = load_image('images/english.jpg').to(torch.bfloat16)
question = '<image>\nPlease Extract text from image'
response_en = model.chat(tokenizer, pixel_values, question, generation_config)

Code

print(response_en)

Write slowly and take the time to make sure each letter is the perfect shape

Chinese Extract

Code

pixel_values = load_image('images/chinese.png').to(torch.bfloat16)

question = '<image>\n提取图上的文字'
response = model.chat(tokenizer, pixel_values, question, generation_config)

Code

print(response)

放养  
把我不羁的灵魂  
放在了鄂西西里的草原上，  
藏雪狐活发爱人  
我与它捉迷藏  
把我不羁的灵魂  
放在撒哈拉沙漠上，  
看生命在贫瘠的土地上，  
依然欣欣欣荣地生长  
把我不羁的灵魂  
放在鲁滨逊的小岛上  
坐上星期五的木筏  
勇敢地乘风破浪  

把我不羁的灵魂  
放养在天涯海角  
让我自由地去流浪。

--- title: "AI图片识别文字" subtitle: "AI Optical character recognition" author: "Tony D" date: "2025-04-21" categories: - AI - R - Python execute: warning: false error: false image: 'images/001.png' --- with Gemini 2.5 online/InternVL3 offline # Using Gemini 2.5 online ```{python} from google import genai import keyring from PIL import Image import torchvision.transforms as T from torchvision.transforms.functional import InterpolationMode import sys import pickle import math import numpy as np import torch ``` ```{python} #| echo: false import pickle f = open('store.pckl', 'rb') response_gemini_en,response_gemini,response_en,response = pickle.load(f) f.close() ``` ```{python} #| eval: false client = genai.Client(api_key=keyring.get_password("system", "google_ai_api_key")) ``` list model ```{python} #| eval: false print("List of models:\n") for m in client.models.list(): for action in m.supported_actions: # if action == "generateContent": print(m.name+" "+ action) ``` ## English Extract ![](images/english.jpg) ```{python} #| eval: false image = Image.open("images/english.jpg") response_gemini_en = client.models.generate_content( model="gemini-2.5-pro-exp-03-25", contents=[image, "Extract text from image"]) ``` ```{python} print(response_gemini_en.text) ``` ## chinese Extract ![](images/chinese.png) ```{python} #| eval: false image = Image.open("images/chinese.png") response_gemini = client.models.generate_content( model="gemini-2.5-pro-exp-03-25", contents=[image, "提取图上的文字"]) ``` ```{python} print(response_gemini.text) ``` # Using InternVL3 1B model offline https://huggingface.co/OpenGVLab/InternVL3-8B-hf If using better model will increase accuracy ```{python} #| eval: false print(sys.version) ``` ```{python} #| eval: false pip install --upgrade transformers pip install einops timm pip install -U bitsandbytes ``` ```{python} #| eval: false import os os.system('pip show transformers') ``` ```{python} #| eval: false import torch from transformers import AutoTokenizer, AutoModel,pipeline path = "OpenGVLab/InternVL3-1B" model = AutoModel.from_pretrained(path,torch_dtype=torch.bfloat16,trust_remote_code=True).eval() tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True) ``` ```{python} generation_config = dict(max_new_tokens=1024, do_sample=True) ``` testing ```{python} #| eval: false question = 'Hello, who are you?' response, history = model.chat(tokenizer, None, question, generation_config, history=None, return_history=True) print(f'User: {question}\nAssistant: {response}') ``` ## single-image single-round conversation (单图单轮对话) ```{python} #| eval: false IMAGENET_MEAN = (0.485, 0.456, 0.406) IMAGENET_STD = (0.229, 0.224, 0.225) def build_transform(input_size): MEAN, STD = IMAGENET_MEAN, IMAGENET_STD transform = T.Compose([ T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img), T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC), T.ToTensor(), T.Normalize(mean=MEAN, std=STD) ]) return transform def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size): best_ratio_diff = float('inf') best_ratio = (1, 1) area = width * height for ratio in target_ratios: target_aspect_ratio = ratio[0] / ratio[1] ratio_diff = abs(aspect_ratio - target_aspect_ratio) if ratio_diff < best_ratio_diff: best_ratio_diff = ratio_diff best_ratio = ratio elif ratio_diff == best_ratio_diff: if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]: best_ratio = ratio return best_ratio def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False): orig_width, orig_height = image.size aspect_ratio = orig_width / orig_height # calculate the existing image aspect ratio target_ratios = set( (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if i * j <= max_num and i * j >= min_num) target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1]) # find the closest aspect ratio to the target target_aspect_ratio = find_closest_aspect_ratio( aspect_ratio, target_ratios, orig_width, orig_height, image_size) # calculate the target width and height target_width = image_size * target_aspect_ratio[0] target_height = image_size * target_aspect_ratio[1] blocks = target_aspect_ratio[0] * target_aspect_ratio[1] # resize the image resized_img = image.resize((target_width, target_height)) processed_images = [] for i in range(blocks): box = ( (i % (target_width // image_size)) * image_size, (i // (target_width // image_size)) * image_size, ((i % (target_width // image_size)) + 1) * image_size, ((i // (target_width // image_size)) + 1) * image_size ) # split the image split_img = resized_img.crop(box) processed_images.append(split_img) assert len(processed_images) == blocks if use_thumbnail and len(processed_images) != 1: thumbnail_img = image.resize((image_size, image_size)) processed_images.append(thumbnail_img) return processed_images def load_image(image_file, input_size=448, max_num=12): image = Image.open(image_file).convert('RGB') transform = build_transform(input_size=input_size) images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num) pixel_values = [transform(image) for image in images] pixel_values = torch.stack(pixel_values) return pixel_values def split_model(model_name): device_map = {} world_size = torch.cuda.device_count() config = AutoConfig.from_pretrained(model_path, trust_remote_code=True) num_layers = config.llm_config.num_hidden_layers # Since the first GPU will be used for ViT, treat it as half a GPU. num_layers_per_gpu = math.ceil(num_layers / (world_size - 0.5)) num_layers_per_gpu = [num_layers_per_gpu] * world_size num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.5) layer_cnt = 0 for i, num_layer in enumerate(num_layers_per_gpu): for j in range(num_layer): device_map[f'language_model.model.layers.{layer_cnt}'] = i layer_cnt += 1 device_map['vision_model'] = 0 device_map['mlp1'] = 0 device_map['language_model.model.tok_embeddings'] = 0 device_map['language_model.model.embed_tokens'] = 0 device_map['language_model.output'] = 0 device_map['language_model.model.norm'] = 0 device_map['language_model.model.rotary_emb'] = 0 device_map['language_model.lm_head'] = 0 device_map[f'language_model.model.layers.{num_layers - 1}'] = 0 return device_map ``` ### English Extract ![](images/english.jpg) ```{python} #| eval: false pixel_values = load_image('images/english.jpg').to(torch.bfloat16) question = '<image>\nPlease Extract text from image' response_en = model.chat(tokenizer, pixel_values, question, generation_config) ``` ```{python} print(response_en) ``` ### Chinese Extract ![](images/chinese.png) ```{python} #| eval: false pixel_values = load_image('images/chinese.png').to(torch.bfloat16) question = '<image>\n提取图上的文字' response = model.chat(tokenizer, pixel_values, question, generation_config) ``` ```{python} print(response) ``` ```{python} #| echo: false # save result f = open('store.pckl', 'wb') pickle.dump([response_gemini_en,response_gemini,response_en,response] ,f) f.close() ```