This is the midterm assignment of DATA130051 Computer Vision.
In the first part of the project, we use several CNNs to build classifiers for CIFAR-100, and try to use some augmentation methods like cutmix, cutout, and mixup to improve the model’s accuracy. And we finally get a top-1 accuracy of 83.01% and top-5 accuracy of 95.70% using the WideResNet-28.
In the second part of the project, we train VOC 2007 using Faster R-CNN and YOLO v3, by the help of pretrained models, we finally gain the mAP0.5 of 77.44% and 86.83% respectively, we also test both models on several images that do not belong to the VOC dataset and get promising results.
The usages of the codes like training and testing processes are shown in the corresponding folders.