Mindspore Image Classifier

Classifying images in the CIFAR-10 dataset

Comment

Inspiration

What it does

It is a model that is able to identify and assign a given set of images with appropriate labels. As we have utilised the CIFAR-10 dataset, the 10 categories that our model is able to classify are:

  • airplane
  • automobile
  • bird
  • cat
  • deer
  • dog
  • frog
  • horse
  • ship
  • truck

How we built it

The entire project was built in Python, making use of the Mindspore Python Module/Library. We had to read through many pages of the documentation, follow tutorials and go through Mindspore's Github repository to be able to produce the result we have submitted.

Challenges we ran into

Going into this project with only some foundational knowledge of ML/AI, the task was a little more advanced than we had intially assumed. The two biggest challenge we faced were:

  • Getting a good grasp and understanding of Mindspore in general and how we are able to incoporate it into the project via code

  • Exploring many of the various image classifying models available and finally deciding on one + figuring out ways to optimise it in terms of accuracy and speed

Accomplishments that we're proud of

As we were beginners in using Mindspore, we faced several tough challenges that was stated above. However, we are proud to make a lot of progress in the short span of time we were given to use this open-sourced inference/training framework. some of the accomplishments we had was:

  • Being able to implement a network (Resnet) to train the model, despite hardware limitation.

  • Enriching our knowledge on machine learning and the Mindspore framework.

What we learned

  • Deeper understanding of the different criteria that needs to be satisfied when creating an image classification model.

  • Find different ways to optimise and enhance the model to produce more accurate results.

  • Proper debugging and evaluation of a image classification model

  • Visualising the training process of the model by creating summary records

  • Explaining how the model works through benchmarking and different explanation methods

What's next for Mindspore Image Classifier

With this project, we envision further improvements for this project such as:

  • Implementing optimisation and general improvements into the code to improve the accuracy and efficiency of the model.

  • Use larger training sets to also improve the accuracy of the model.

Built With

Share this project:

Updates