vlmtest/models/LLaVA_Video.py

# pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
from llava.model.builder import load_pretrained_model
from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX
from llava.conversation import conv_templates, SeparatorStyle
from PIL import Image
import requests
import copy
import torch
import sys
import warnings
from decord import VideoReader, cpu
import numpy as np
warnings.filterwarnings("ignore")

def load_images(images_path):
    images = []
    for image_path in images_path:
        # 打开图像并转换为 NumPy 数组
        image = Image.open(image_path).convert("RGB")  # 确保图像是 RGB 格式
        image_array = np.array(image)
        images.append(image_array)
    
    # 将所有图像组合成一个 NumPy 数组
    images_array = np.stack(images, axis=0)
    return images_array
        
class LLaVA_Video(object):
    def __init__(self, gpu=1, model_path="lmms-lab/LLaVA-Video-7B-Qwen2"):
        self.model_name = "llava_qwen"
        self.device = torch.device(f"cuda:{gpu}" if torch.cuda.is_available() else "cpu")
        self.device_map = {"": f"cuda:{gpu}"}
        self.tokenizer, self.model, self.image_processor, self.max_length = load_pretrained_model(
            model_path, None, 
            self.model_name, torch_dtype="bfloat16", 
            device_map=self.device_map
        )
        self.model.eval()
        
    def inference(self, images_path, qa):
        images = load_images(images_path)
        images = self.image_processor.preprocess(images, return_tensors="pt")["pixel_values"].to(self.device).to(torch.bfloat16)
        images = [images]
        conv_template = "qwen_1_5"  # Make sure you use correct chat template for different models
        question = DEFAULT_IMAGE_TOKEN + f"This question is about the main topic discussed in the video. Question: {qa['question']} Choices: A) {qa['choice_a']} B) {qa['choice_b']} C) {qa['choice_c']} D) {qa['choice_d']}. Respond with a single capital letter (A, B, C, or D) only. No explanation. No punctuation. Just the letter."
        conv = copy.deepcopy(conv_templates[conv_template])
        conv.append_message(conv.roles[0], question)
        conv.append_message(conv.roles[1], None)
        prompt_question = conv.get_prompt()
        input_ids = tokenizer_image_token(prompt_question, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(self.device)
        '''print(type(input_ids))
        print(type(images))'''
        cont = self.model.generate(
            input_ids,
            images=images,
            modalities= ["video"],
            do_sample=False,
            temperature=0,
            max_new_tokens=4096,
        )
        result = self.tokenizer.batch_decode(cont, skip_special_tokens=True)[0].strip()
        return result