Assignment 5 - Course Project

If someone is interested here are my pieces.


Can we use TensorFlow for project?

albumentation has various transformations that are not available in pytorch.

1 Like

I hope this notebook might help you

Yes, it is also because of the F_score. You should not use the F_score as a metrics, instead, use the accuracy. You can refer to the notebook of the fourth lecture (in [16]) it will help you. You can use the same accuracy code and also the ImageClassificationBase model code.

7 posts were merged into an existing topic: Share your work : June 22 - June 28

../input/aerial-cactus-identification/ should work.

These are the mean and standard deviation values for all the ~1.5 Million images in the ImageNet dataset. These are the most commonly used normalisation stats. If you are a begineer, this is a great point to start and these work well with almost any CV problem.

1 Like

@mt21667 Unless you are having test and train sets that are considerably different, this is most probably overfitting. Try using early stopping.

@ankitvashisht12 I think you are missing a sigmoid layer at the end in your architecture.

I am working on a dateset which contains only pixels in a string format as image, Is there any function in PyTorch or python to convert a string of pixels to an image.
can anyone please help me out

Hi! Sorry I forgot to reply. I already solved the problem because of you. I got 98% accuracy rate. Thank you so much! I hope we’ll get in touch again some day.

1 Like

Thanks for your suggestion, i’ll try to use early stopping and see whether it helps my model or not.

Hi everyone,
Anybody know how to upload .pth i.e. saved model to kaggle/working directory ?

I can’t figure out how can i do this :thinking:

Thanks @anirudha-akela :blush:

maximum interactive GPU session count of 1 reached…how to solve this problem

@sathishbabu992 that’s right, you can have only one session running on GPU. check your notebook tab on kaggle, if you have some session running you might want to click ‘Stop interactive session’ first.

1 Like

How many course project can one submit?

@PrajwalPrashanth @aakashns and my fellow coders, quick question:
how do i calculate mean and std dev for grayscale images ?

I am having a string of pixels of a B/W image in this manner:-

48 34 21 18 16 21 26 36 40...44 53 57 64 82 95 9

of string length 547, and I want these pixels to be converted into an image of 48*48 pixel B/W image.

Is there any python library or PyTorch function to make it possible?

Hi All,

I have a data set where for the same client there are about 6 different entries. Should I average the information based on the client?