Upload 226 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+AKSHAYRAJAA/ngrok-v3-stable-linux-amd64.zip.1 filter=lfs diff=lfs merge=lfs -text

AKSHAYRAJAA/README.md

@@ -0,0 +1,52 @@
+# Chest X-Ray Report Generator
+
+> This project is part of a task for the college where I study, so `task-parts` contains files that associated with that task, whishing that I would get the full mark ;). In general the base code doesn't have any special parts except that folder.
+
+## Installation
+
+After cloning the repository, install the required packages in a virtual environment.
+
+Next, download the datasets and checkpoints, as describe below.
+
+## Dataset
+
+### IU X-Ray
+
+1. Download the Chen et al. labels and the chest X-rays in png format for IU X-Ray from:
+
+```
+https://openi.nlm.nih.gov
+```
+
+2. Place the files into `dataset` folder, such that their paths are `dataset/reports` and `dataset/images`.
+
+## Checkpoints
+
+This approach uses `CheXNet`, and `DenseNet121` as a CNN Encoder model. By default the `CheXNet` pretrained weights are located in `weights` folder.
+
+## Config
+
+The model configurations for each task can be found in its `config.py` file.
+
+## Training and Evaluation
+
+### Training
+
+Use the below command to train the model form a saved checkpoint or without a checkpoint.
+
+```bash
+python train.py
+```
+
+### Evaluation
+
+The model performance measure is based of the `BLEU` metric.
+
+> Feel free to change the performance measure metric in the `check_accuracy` method that is located in the `eval.py` file
+
+Run the following command to calculate `BLEU` score.
+
+```bash
+python eval.py
+```
+

AKSHAYRAJAA/checkpoints.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ed4226fab578a672602a194b066fdeb3b72225c3054b955fcdba2e8b59cb661
+size 65840736

AKSHAYRAJAA/checkpoints/x_ray_model.pth.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aca89cd72ca242bf8e2e8406ccd855fcdd922fe788583eb109af765100b169be
+size 65840564

AKSHAYRAJAA/config.py

@@ -0,0 +1,36 @@
+import albumentations as A
+import torch
+
+from albumentations.pytorch import ToTensorV2
+
+
+CHECKPOINT_FILE = 'd:\\AKSHAYRAJAA\\checkpoints\\x_ray_model.pth.tar'
+DATASET_PATH = 'D:\\AKSHAYRAJAA\\dataset\\'
+IMAGES_DATASET = 'D:\\AKSHAYRAJAA\\dataset\\images'
+
+DEVICE = 'cpu'
+BATCH_SIZE = 16
+PIN_MEMORY = False
+VOCAB_THRESHOLD = 2
+
+FEATURES_SIZE = 1024
+EMBED_SIZE = 300
+HIDDEN_SIZE = 256
+
+LEARNING_RATE = 4e-5
+EPOCHS = 50
+
+LOAD_MODEL = True
+SAVE_MODEL = True
+
+basic_transforms = A.Compose([
+    A.Resize(
+        height=256,
+        width=256
+    ),
+    A.Normalize(
+        mean=(0.485, 0.456, 0.406),
+        std=(0.229, 0.224, 0.225),
+    ),
+    ToTensorV2()
+])

AKSHAYRAJAA/dataset.py

@@ -0,0 +1,165 @@
+import os
+import spacy
+import torch
+import config
+import utils
+import numpy as np
+import xml.etree.ElementTree as ET
+
+from PIL import Image
+from torch.nn.utils.rnn import pad_sequence
+from torch.utils.data import Dataset, DataLoader
+
+
+spacy_eng = spacy.load('en_core_web_sm')
+
+
+class Vocabulary:
+    def __init__(self, freq_threshold):
+        self.itos = {
+            0: '<PAD>',
+            1: '<SOS>',
+            2: '<EOS>',
+            3: '<UNK>',
+        }
+        self.stoi = {
+            '<PAD>': 0,
+            '<SOS>': 1,
+            '<EOS>': 2,
+            '<UNK>': 3,
+        }
+        self.freq_threshold = freq_threshold
+
+    @staticmethod
+    def tokenizer(text):
+        return [tok.text.lower() for tok in spacy_eng.tokenizer(text)]
+
+    def build_vocabulary(self, sentence_list):
+        frequencies = {}
+        idx = 4
+
+        for sent in sentence_list:
+            for word in self.tokenizer(sent):
+                if word not in frequencies:
+                    frequencies[word] = 1
+                else:
+                    frequencies[word] += 1
+
+                if frequencies[word] == self.freq_threshold:
+                    self.stoi[word] = idx
+                    self.itos[idx] = word
+
+                    idx += 1
+
+    def numericalize(self, text):
+        tokenized_text = self.tokenizer(text)
+
+        return [
+            self.stoi[token] if token in self.stoi else self.stoi['<UNK>']
+            for token in tokenized_text
+        ]
+
+    def __len__(self):
+        return len(self.itos)
+
+
+class XRayDataset(Dataset):
+    def __init__(self, root, transform=None, freq_threshold=3, raw_caption=False):
+        self.root = root
+        self.transform = transform
+        self.raw_caption = raw_caption
+
+        self.vocab = Vocabulary(freq_threshold=freq_threshold)
+
+        self.captions = []
+        self.imgs = []
+
+        for file in os.listdir(os.path.join(self.root, 'reports')):
+            if file.endswith('.xml'):
+                tree = ET.parse(os.path.join(self.root, 'reports', file))
+
+                frontal_img = ''
+                findings = tree.find(".//AbstractText[@Label='FINDINGS']").text
+
+                if findings is None:
+                    continue
+
+                for x in tree.findall('parentImage'):
+                    if frontal_img != '':
+                        break
+
+                    img = x.attrib['id']
+                    img = os.path.join(config.IMAGES_DATASET, f'{img}.png')
+
+                    frontal_img = img
+
+                if frontal_img == '':
+                    continue
+                
+                self.captions.append(findings)
+                self.imgs.append(frontal_img)
+                
+
+        self.vocab.build_vocabulary(self.captions)
+
+    def __getitem__(self, item):
+        img = self.imgs[item]
+        caption = utils.normalize_text(self.captions[item])
+
+        img = np.array(Image.open(img).convert('L'))
+        img = np.expand_dims(img, axis=-1)
+        img = img.repeat(3, axis=-1)
+
+        if self.transform is not None:
+            img = self.transform(image=img)['image']
+
+        if self.raw_caption:
+            return img, caption
+        
+        numericalized_caption = [self.vocab.stoi['<SOS>']]
+        numericalized_caption += self.vocab.numericalize(caption)
+        numericalized_caption.append(self.vocab.stoi['<EOS>'])
+
+        return img, torch.as_tensor(numericalized_caption, dtype=torch.long)
+
+    def __len__(self):
+        return len(self.captions)
+
+    def get_caption(self, item):
+        return self.captions[item].split(' ')
+
+
+class CollateDataset:
+    def __init__(self, pad_idx):
+        self.pad_idx = pad_idx
+
+    def __call__(self, batch):
+        images, captions = zip(*batch)
+
+        images = torch.stack(images, 0)
+        
+        targets = [item for item in captions]
+        targets = pad_sequence(targets, batch_first=True, padding_value=self.pad_idx)
+
+        return images, targets
+
+
+if __name__ == '__main__':
+    all_dataset = XRayDataset(
+        root=config.DATASET_PATH,
+        transform=config.basic_transforms,
+        freq_threshold=config.VOCAB_THRESHOLD,
+    )
+
+    train_loader = DataLoader(
+        dataset=all_dataset,
+        batch_size=config.BATCH_SIZE,
+        pin_memory=config.PIN_MEMORY,
+        drop_last=True,
+        shuffle=True,
+        collate_fn=CollateDataset(pad_idx=all_dataset.vocab.stoi['<PAD>']),
+    )
+
+    for img, caption in train_loader:
+        print(img.shape, caption.shape)
+        break