mousie
Reputation: 21
I don't know why I'm receiving "Exception has occurred" in this situation
I'm trying to run code from https://github.com/DDI-Dataset/DDI-Code with its dataset (already downloaded and added to a folder, as instructed by the author), and I get the message "Exception has occurred: Exception Please visit https://stanfordaimi.azurewebsites.net/datasets/35866158-8196-48d8-87bf-50dca81df965 to download the DDI dataset." while using the python debugger in VScode.
Below is the code from the file ddi_dataset.py.
"""Code for loading DDI Dataset."""
from torch.utils.data import Subset
from torchvision.datasets import ImageFolder
from torchvision import transforms as T
import os
import pandas as pd
import numpy as np
means = [0.485, 0.456, 0.406]
stds = [0.229, 0.224, 0.225]
test_transform = T.Compose([
lambda x: x.convert('RGB'),
T.Resize(299),
T.CenterCrop(299),
T.ToTensor(),
T.Normalize(mean=means, std=stds)
])
class DDI_Dataset(ImageFolder):
_DDI_download_link = "https://stanfordaimi.azurewebsites.net/datasets/35866158-8196-48d8-87bf-50dca81df965"
"""DDI Dataset.
Args:
root (str): Root directory of dataset.
csv_path (str): Path to the metadata CSV file. Defaults to `{root}/ddi_metadata.csv`
transform : Function to transform and collate image input. (can use test_transform from this file)
"""
# root = 'C:\\Users\\User\\OneDrive\\desktop\\CSproject\\datasets\\DDI\\images'
def __init__(self, root, csv_path=None, download=True, transform=None, *args, **kwargs):
if csv_path is None:
csv_path = os.path.join(root, "ddi_metadata.csv")
if not os.path.exists(csv_path) and download:
raise Exception(f"Please visit <{DDI_Dataset._DDI_download_link}> to download the DDI dataset.")
assert os.path.exists(csv_path), f"Path not found <{csv_path}>."
super(DDI_Dataset, self).__init__(root, *args, transform=transform, **kwargs)
self.annotations = pd.read_csv(csv_path)
m_key = 'malignant'
if m_key not in self.annotations:
self.annotations[m_key] = self.annotations['malignancy(malig=1)'].apply(lambda x: x==1)
def __getitem__(self, index):
img, target = super(DDI_Dataset, self).__getitem__(index)
path = self.imgs[index][0]
annotation = dict(self.annotations[self.annotations.DDI_file==path.split("/")[-1]])
target = int(annotation['malignant'].item()) # 1 if malignant, 0 if benign
skin_tone = annotation['skin_tone'].item() # Fitzpatrick- 12, 34, or 56
return path, img, target, skin_tone
"""Return a subset of the DDI dataset based on skin tones and malignancy of lesion.
Args:
skin_tone (list of int): Which skin tones to include in the subset. Options are {12, 34, 56}.
diagnosis (list of str): Include malignant and/or benign images. Options are {"benign", "malignant"}
"""
def subset(self, skin_tone=None, diagnosis=None):
skin_tone = [12, 34, 56] if skin_tone is None else skin_tone
diagnosis = ["benign", "malignant"] if diagnosis is None else diagnosis
for si in skin_tone:
assert si in [12,34,56], f"{si} is not a valid skin tone"
for di in diagnosis:
assert di in ["benign", "malignant"], f"{di} is not a valid diagnosis"
indices = np.where(self.annotations['skin_tone'].isin(skin_tone) & \
self.annotations['malignant'].isin([di=="malignant" for di in diagnosis]))[0]
return Subset(self, indices)
I thought that the problem is that the variable root does not include any file path so I tried to add an absolute file path to it (the comment above, root = 'C:\\Users\\User\\OneDrive\\desktop\\CSproject\\datasets\\DDI\\images'); however, the same error message came out.
Below is the file's location in my computer.
ddi_dataset.py: C:\Users\User\OneDrive\desktop\CSproject\dataset\DDI\ddi_dataset.py
Images:
C:\Users\User\OneDrive\desktop\CSproject\dataset\DDI\images
('images' is the folder where I put the PNG:s and ddi_metadata.csv)
Additionally, below is the code from eval_ddi.py where it calls the DDI_Dataset
"""
Code to evaluate the models trained for our paper,
"Disparities in Dermatology AI Performance on a Diverse,
Curated Clinical Image Set",
on the DDI dataset.
Note: assumes DDI data is organized as
./DDI
/images
/000001.png
/000002.png
...
/ddi_metadata.csv
(After downloading from the Stanford AIMI repository, this requires moving all .png files into a new subdirectory titled "images".)
------------------------------------------------------
Examples:
(1) w/command line interface
# evaluate DeepDerm on DDI and store results in `DDI-results`
>>>python3 eval_ddi.py --model=DeepDerm --data_dir=DDI --eval_dir=DDI-results
(2) w/python functions
>>>import eval_ddi
>>>import ddi_model
>>>model = ddi_model.load_model("DeepDerm") # load DeepDerm model
>>>eval_results = eval_ddi.eval_model(model, "DDI") # evaluate images in DDI folder
"""
import argparse
from ddi_dataset import DDI_Dataset, test_transform
from ddi_model import load_model
import matplotlib.pyplot as plt
import numpy as np
import os
import pickle
from sklearn.metrics import (f1_score, balanced_accuracy_score,
classification_report, confusion_matrix, roc_curve, auc)
import torch
import tqdm
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument('--model_dir', type=str, default="DDI-models",
help="File path for where to save models.")
parser.add_argument('--model', type=str, default="DeepDerm",
help="Name of the model to load (HAM10000, DeepDerm, GroupDRO, CORAL,"\
" or CDANN).")
parser.add_argument('--no_download', action='store_true', default=False,
help="Set to disable downloading models.")
parser.add_argument('--data_dir', type=str, default="DDI",
help="Folder containing dataset to load. Structure should be: (1) `[data_dir]/images` contains all images; (2) `[data_dir]/ddi_metadata.csv` contains the CSV metadata for the DDI dataset")
parser.add_argument('--eval_dir', type=str, default="DDI-results",
help="Folder to store evaluation results.")
parser.add_argument('--use_gpu', action='store_true', default=False,
help="Set to use GPU for evaluation.")
parser.add_argument('--plot', action='store_true', default=False,
help="Set to show ROC plot.")
args = parser.parse_args()
return args
def eval_model(model, dataset, use_gpu=False, show_plot=False):
"""Evaluate loaded model on provided image dataset. Assumes supplied image
directory corresponds to `root` input for torchvision.datasets.ImageFolder
class. Assumes the data is split into binary/malignant labels, as this is
what our models are trained+evaluated on."""
use_gpu = (use_gpu and torch.cuda.is_available())
device = torch.device("cuda") if use_gpu else torch.device("cpu")
# load dataset
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=32, shuffle=False,
num_workers=0, pin_memory=use_gpu)
# prepare model for evaluation
model.to(device).eval()
# log output for all images in dataset
hat, star, all_paths = [], [], []
for batch in tqdm.tqdm(enumerate(dataloader)):
i, (paths, images, target, skin_tone) = batch
images = images.to(device)
target = target.to(device)
with torch.no_grad():
output = model(images)
hat.append(output[:,1].detach().cpu().numpy())
star.append(target.cpu().numpy())
all_paths.append(paths)
hat = np.concatenate(hat)
star = np.concatenate(star)
all_paths = np.concatenate(all_paths)
threshold = model._ddi_threshold
m_name = model._ddi_name
m_web_path = model._ddi_web_path
report = classification_report(star, (hat>threshold).astype(int),
target_names=["benign","malignant"])
fpr, tpr, _ = roc_curve(star, hat, pos_label=1,
sample_weight=None,
drop_intermediate=True)
auc_est = auc(fpr, tpr)
if show_plot:
_=plt.plot(fpr, tpr,
color="blue", linestyle="-", linewidth=2,
marker="o", markersize=2,
label=f"AUC={auc_est:.3f}")[0]
plt.show()
plt.close()
eval_results = {'predicted_labels':hat, # predicted labels by model
'true_labels':star, # true labels
'images':all_paths, # image paths
'report':report, # sklearn classification report
'ROC_AUC':auc_est, # ROC-AUC
'threshold':threshold, # >= threshold ==> malignant
'model':m_name, # model name
'web_path':m_web_path, # web link to download model
}
return eval_results
if __name__ == '__main__':
# get arguments from command line
args = get_args()
# load model and download if necessary
model = load_model(args.model,
save_dir=args.model_dir, download=not args.no_download)
# load DDI dataset
dataset = DDI_Dataset("DDI", transform=test_transform)
# evaluate results on data
eval_results = eval_model(model, dataset,
use_gpu=args.use_gpu, show_plot=args.plot)
# save evaluation results in a .pkl file
if args.eval_dir:
os.makedirs(args.eval_dir, exist_ok=True)
eval_save_path = os.path.join(args.eval_dir,
f"{args.model}-evaluation.pkl")
with open(eval_save_path, 'wb') as f:
pickle.dump(eval_results, f)
# load results with:
#with open(eval_save_path, 'rb') as f:
# results = pickle.load(f)
and here is the full trace back:
Exception has occurred: Exception
Please visit <https://stanfordaimi.azurewebsites.net/datasets/35866158-8196-48d8-87bf-50dca81df965> to download the DDI dataset.
File "C:\Users\User\OneDrive\desktop\CSproject\dataset\DDI\ddi_dataset.py", line 37, in __init__
raise Exception(f"Please visit <{DDI_Dataset._DDI_download_link}> to download the DDI dataset.")
File "C:\Users\User\OneDrive\desktop\CSproject\dataset\DDI\eval_ddi.py", line 140, in <module>
dataset = DDI_Dataset("DDI", transform=test_transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Exception: Please visit <https://stanfordaimi.azurewebsites.net/datasets/35866158-8196-48d8-87bf-50dca81df965> to download the DDI dataset.
I'm expecting to know why the problem happens and how to solve it.
Upvotes: 0
Views: 91
Answers (1)
tripleee
Reputation: 189789
The file ddi_metadata.csv must be in the root folder, not in images
Upvotes: 0