Pytorch

DCGAN

• dcgan.ipynb

  • This notebook demonstrates this process on the Tachikara basketball custom dataset.
  • The following animation shows a series of images produced by the generator as it was trained
  
  • Real Image vs Fake Image
  

Image Classification

Using VGG & Resnet in PyTorch torchvision.models, and train CIFAR10 dataset for image classification

• image_classification.ipynb

  • the notebook is using GPU on Google colab

• Learning rate adjustment by StepLR from torch.optim.lr_scheduler

• Result

  • Accuracy

    Model	Accuracy
    VGG16	85 %
    ResNet18	80 %

  • Accuracy for each class

    	VGG16	Resnet18
    plane	82 %	86 %
    car	100 %	85 %
    bird	90 %	75 %
    cat	58 %	58 %
    deer	85 %	88 %
    dog	84 %	72 %
    frog	86 %	83 %
    horse	76 %	92 %
    ship	96 %	96 %
    truck	89 %	87 %

Object Detection

Using Pre-trained Faster RCNN in PyTorch torchvision.models.detection for object detection

• object_detection_image.ipynb
  • the notebook is for image and the workflow for the detection
• object_detection_video.py
  • object detection for the video

Image	Video

Instance Segmentation

Using Pre-trained Mask RCNN in PyTorch torchvision.models.detection for instance segmentation

• instance_segmentation_image.ipynb
  • the notebook is for image and the workflow for the instance_segmentation
• instance_segmentation_video.py
  • instance_segmentation for the video

Image	Video

Reference

• Faster R-CNN Tutorial
• Classfier Tutorial
• Instance-Segmentation Tutorial
• custom-trained-object-detection-model