人工智能与机器学习 作业 ¶

rm -rf main.ipynb
rm -rf results/

zip -r main.zip ./*

unzip main.zip

scp main.zip <username>@<ip>:<path>

scp <username>@<ip>:<path> <path> 

image = Image.fromarray(np.array(image)[:, :, ::-1])

import torch
from PIL import Image
import torchvision.transforms as transforms
import torch.nn as nn
import os
import numpy as np
os.system(f'pip install timm') # 注意这里需要安装timm
os.system(f'pip install ttach')
import timm
import ttach as tta
inverted = {
    0: 'Plastic Bottle', 1: 'Hats', 2: 'Newspaper', 3: 'Cans', 4: 'Glassware', 5: 'Glass Bottle',
    6: 'Cardboard', 7: 'Basketball', 8: 'Paper', 9: 'Metalware', 10: 'Disposable Chopsticks',
    11: 'Lighter', 12: 'Broom', 13: 'Old Mirror', 14: 'Toothbrush', 15: 'Dirty Cloth', 16: 'Seashell',
    17: 'Ceramic Bowl', 18: 'Paint bucket', 19: 'Battery', 20: 'Fluorescent lamp', 21: 'Tablet capsules',
    22: 'Orange Peel', 23: 'Vegetable Leaf', 24: 'Eggshell', 25: 'Banana Peel'
}

# 全局，加载模型，避免每一次都加载浪费时间
device = 'cuda' if torch.cuda.is_available() else 'cpu'
model_1 = timm.create_model('seresnext50_32x4d.racm_in1k', pretrained=False)
model_1.fc = nn.Linear(model_1.num_features, 26)
model_1.load_state_dict(torch.load('model_sere.pth', map_location=device)['state_dict'])
model_1 = model_1.to(device)


model_2 = timm.create_model('seresnext50_32x4d.racm_in1k', pretrained=False)
model_2.fc = nn.Linear(model_2.num_features, 26)
model_2.load_state_dict(torch.load('model2.pth', map_location=device)['state_dict'])
model_2 = model_2.to(device)

def predict(image):
    # 如果输入是numpy数组,转换为PIL Image
    if isinstance(image, np.ndarray):
        image = Image.fromarray(image)
    # 确保图像是RGB模式
    if image.mode != 'RGB':
        image = image.convert('RGB')
    image = Image.fromarray(np.array(image)[:, :, ::-1])
    transform = transforms.Compose([
        transforms.Resize(224),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # 这个是imagenet的均值和方差
    ])
    imgs = transform(image).unsqueeze(0).to(device)
    
    model_1.eval()
    model_2.eval()

    tta_model_1 = tta.ClassificationTTAWrapper(model_1, tta.aliases.flip_transform(), merge_mode='mean')
    tta_model_2 = tta.ClassificationTTAWrapper(model_2, tta.aliases.flip_transform(), merge_mode='mean')

    with torch.no_grad():
        logits_1 = tta_model_1(imgs.to(device))  #预测
        logits_2 = tta_model_2(imgs.to(device))
        logits = logits_1 + logits_2 
        return inverted[logits.argmax(dim=-1).cpu().numpy().item()]

# 之后整理一下再放

import torch
import torch.nn as nn
from transformers import BertModel, BertTokenizer

# 通过torch.hub(pytorch中专注于迁移学的工具)获得已经训练好的bert-base-chinese模型
# model =  torch.hub.load('huggingface/pytorch-transformers', 'model', 'bert-base-chinese')
model = BertModel.from_pretrained('D:\\MyPython\\data\\bert-base-chinese')

# 获得对应的字符映射器, 它将把中文的每个字映射成一个数字
# tokenizer = torch.hub.load('huggingface/pytorch-transformers', 'tokenizer', 'bert-base-chinese')
tokenizer = BertTokenizer.from_pretrained('D:\\MyPython\\data\\bert-base-chinese')