Unable to get a prediction accuracy

I have used intel-image classification.
In this we have three folder.

1.Train folder
2.Test folder
3.Pred folder

I have used train folder for training and and test folder for validation. Now when i am prediction on pred folder im not getting desired accuray.
test folder used for validation-------acuuracy is 77%
pred folder used for prediction-----------accuracy is 9%
Also facing below issue

Please let me know if i am doing something wrong.

Here is my link to notebook.

Hi there, the problem is with your prediction folder i.e. seg_pred, of your dataset.

As you can see, the images in this folder is not subcategorized into different classes. Hence, ImageFolder function is assigning the label 0 to each image and that’s why you are getting the label: buildings for each image.

And same is the reason for getting the prediction accuracy of 9 %.

I will suggest use validation set for also testing your model.

As for images in prediction folder, you can manually see/check, if the model is predicting correctly or not. :slightly_smiling_face:

2 Likes

Thanks @ayush-mishra-ald