https://github.com/tnfru/computer_vision
Classification with Deep Learning
https://github.com/tnfru/computer_vision
cifar10 computer-vision mnist pytorch
Last synced: 6 months ago
JSON representation
Classification with Deep Learning
- Host: GitHub
- URL: https://github.com/tnfru/computer_vision
- Owner: tnfru
- Created: 2020-01-17T14:51:27.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2020-10-07T12:31:18.000Z (almost 6 years ago)
- Last Synced: 2025-02-09T12:24:02.158Z (over 1 year ago)
- Topics: cifar10, computer-vision, mnist, pytorch
- Language: Jupyter Notebook
- Homepage:
- Size: 952 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Computer Vision
In this repo I will upload all problem statements I tackled using computer vison.
I am exectuing these Notebooks on Google Colab, which is a fantastic platform
offering free GPUs and TPUs.
In these I am not using transfer learning as it would basically kill the idea
of learning to design CNN architectures.
## CIFAR-10
We use a Wide-ResNet and achieve 93.23% accuracy, beating the original ResNet56v1
by .2%.
This is due to Dropout and the parallell convolutions in conv and identity modules.
## MNIST
My take on the classic MNIST dataset to classify handwritten digits, which
achieves 99.6% accuracy.
There are a PyTorch and a Keras implementations in this repository.
The Keras version hit 0.99614 on kaggle. It uses a slightly changed version of
LeNet5 replacing 5x5 Filter with two 3x3 for non linerarity.
The PyTorch implementation, which uses a large number of epochs,
checks for overfitting and will abort on its own. This version
was programmed only on a CPU, hence the small network.