'init'

.gitignore

@@ -0,0 +1,19 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+
+# PyBuilder
+target/
+.idea
+dataset/*
+sentence/*
+!.keep
+
+# vector cache
+vector_cache/
+
+# model cache
+model_dir/
+
+# feature cache
+feature_dir/

README.md

@@ -0,0 +1,58 @@
+# PascalSentenceDataset
+
+This program is utility to download pascal sentence dataset.
+
+## Installation
+
+You can install by "git clone" command.
+
+```
+git clone https://github.com/rupy/PascalSentenceDataset.git
+```
+
+### Dependency
+
+You must install some python libraries. Use pip command. Python>=2
+
+```
+PyQuery
+```
+
+## Usage
+
+To download dataset, just run program as follow:
+
+```
+python pascal_sentence_dataset.py
+```
+
+You can also write code like this:
+
+```python
+# import
+from pascal_sentence_dataset import PascalSentenceDataSet
+
+# create instance
+dataset = PascalSentenceDataSet()
+# download images
+dataset.download_images()
+# download sentences
+dataset.download_sentences()
+# create correspondence data by dataset
+# dataset.create_correspondence_data()
+
+# create my pair data
+dataset.create_pair_data()
+# preprocess data
+dataset.preprocess_data()
+```
+
+Return the following file list: (./list/)
+- _correspondence.csv_ 1000 list data, titled: index, image
+- _data_pairs.csv_ 1000 list data, titled: index, image, text, label
+- _train.csv_ the training set with 800 image-text pairs (40 pairs per class)
+- _validate.csv_ 100 the validation set with 100 image-text pairs (5 pairs per class)
+- _test.csv_ 100 the testing set with 100 image-text pairs (5 pairs per class)
+
+
+

dataset.py

@@ -0,0 +1,65 @@
+import torch
+
+import torch.nn as nn
+from torchvision import transforms
+from PIL import Image
+from torchtext import data
+from torchtext.data import get_tokenizer
+from torch.nn import init
+
+
+
+
+
+# VGG模型提取图像特征
+class VGG:
+    def __init__(self):
+        super().__init__()
+
+        # load image model
+        image_model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg19', pretrained=True)
+        # or any of these variants
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg11_bn', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg13', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg13_bn', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg16', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg16_bn', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg19', pretrained=True)
+        # model = torch.hub.load('pytorch/vision:v0.6.0', 'vgg19_bn', pretrained=True)
+
+        image_model.eval()  # 处于验证状态
+        image_model.classifier._modules['6'] = nn.Identity()  # update the last layer
+        self.image_model = image_model.to(DEVICE)
+
+        # set image transform
+        self.preprocess = transforms.Compose([
+            transforms.Resize(256),
+            transforms.CenterCrop(224),
+            transforms.ToTensor(),
+            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+        ])
+
+    # 裁剪图片，输出为1×3×224×224
+    def transform_image(self, image):
+        input_tensor = self.preprocess(image)
+        return input_tensor.unsqueeze(0)  # create a mini-batch as expected by the model
+
+    # 获取图片特征，输出为4096维
+    def feature_image(self, batch):
+        batch = batch.to(DEVICE)
+        with torch.no_grad():
+            return self.image_model(batch)
+
+
+# 获取图像的VGG特征
+def get_image_tokenize(image_file):
+    input_image = Image.open('./dataset/' + image_file)
+    input_batch = vgg19.transform_image(input_image)
+    return vgg19.feature_image(input_batch)
+
+
+# 可用设备
+DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# 初始化VGG模型
+vgg19 = VGG()