Scribd Logo
Upload

Welcome to Scribd!

  • CV Assignment - Object Recognition Using CNN

    Uploaded by

    Danyal Ahmed
    23 views1 page
    This document discusses using deep learning architectures for object recognition. It describes training convolutional networks from scratch or fine-tuning pre-trained networks for recognition using PyTorch and TensorFlow. The document instructs to run small CNNs on MNIST and CIFAR10 datasets, tune hyperparameters, and observe the effects of different choices. It also describes adding preprocessing, dropout, batch normalization and more layers to a simple network, and finally fine-tuning a pre-trained network.

    Original Description:

    Original Title

    CV assignment - object recognition using CNN

    Copyright

    © © All Rights Reserved

    Available Formats

    DOCX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    This document discusses using deep learning architectures for object recognition. It describes training convolutional networks from scratch or fine-tuning pre-trained networks for recognition using PyTorch and TensorFlow. The document instructs to run small CNNs on MNIST and CIFAR10 datasets, tune hyperparameters, and observe the effects of different choices. It also describes adding preprocessing, dropout, batch normalization and more layers to a simple network, and finally fine-tuning a pre-trained network.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    23 views1 page

    CV Assignment - Object Recognition Using CNN

    Uploaded by

    Danyal Ahmed
    This document discusses using deep learning architectures for object recognition. It describes training convolutional networks from scratch or fine-tuning pre-trained networks for recognition using PyTorch and TensorFlow. The document instructs to run small CNNs on MNIST and CIFAR10 datasets, tune hyperparameters, and observe the effects of different choices. It also describes adding preprocessing, dropout, batch normalization and more layers to a simple network, and finally fine-tuning a pre-trained network.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    of 1

    Object recognition using deep learning architectures

    This assignment gives you practice with the prevalent tools in object recognition. You will design and
    train deep convolutional networks for recognition using PyTorch, following up with the tutorial in TF that
    you saw in the sessions.

    These different approaches (starting the training from scratch or fine-tuning) represent the most
    common approach to recognition problems in computer vision today -- train a deep network from
    scratch if you have enough data (it's not always obvious whether or not you do), and if you cannot then
    fine-tune a pre-trained network instead.

    You would run small CNNs on 1. MNIST, and 2. CIFAR10. These examples are already present in the PT
    online doc. You have to tune your hyperparameters to achieve different results and observe what
    changes when you prefer certain hyperparameter choices.

    First you will train a simple network from scratch. Report the performance of this simple network.

    Then, you will modify the dataloader to include a few extra pre-processing steps. Pre-processing data
    allows the network to be more robust against small variations in the data. Examples of pre-processing
    transforms are jittering, flipping, and normalization. You will also modify the simple network by adding
    dropout, batch normalizations and more layers. Report your findings.

    In the end for your curiosity, you may instead fine-tune a pre-trained deep network to achieve more
    accuracy on the task.

    You might also like