roykesydon/cvdl-homework-1

CVDL-Homework-1

Objective

1. Camera calibration

   • Find chessboard corners
   • Find the intrinsic, extrinsic and distortion parameters of the camera
   • Use the parameters to undistort the image

2. Augmented reality

   • Transform real world coordinates to image coordinates
   • Draw given word on the image

3. Stereo Disparity Map

   • calculate the disparity map of the given stereo image pair
   • draw corresponding points on the image pair according to the disparity map

4. SIFT

   • Find the keypoints and descriptors of the given image
   • Match the keypoints of the given image pair

5. Train VGG19_bn on CIFAR10

Environment

• Python 3.8

Setup

1. Install dependencies

   pip install -r requirements.txt --no-cache-dir

   Windows 要裝有 CUDA 的 Pytorch 再跑以下指令

   pip3 install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

   MacOS 裝 PyQt5 遇到問題的話，建議用 conda 裝

   conda install -c anaconda pyqt

2. Check whether torch is using GPU

   python test_gpu.py

3. Put pretrained weights vgg_16_bn_final.pth in ./weights/

Run

python app.py

Train

After training, the weights will be saved in ./weights/ and the result chart will be saved in ./result.png

python train.py