Fine-grained detection on Vehicle Model/Make
Training dataset consisted of 163/1,716 vehicle make/models from CompCars dataset[1]
A VGG16 model pre-trained on ImageNet was fine-tuned with CompCars dataset (16970/776 train/valid images - 115 vehicles/classes)
Accuracy: 93.2% top-5 in 200 epochs Base learning rate of 0.001 and batch size of 64 were used.
[1] Linjie Yang, Ping Luo, Chen Change Loy, Xiaoou Tang. A Large-Scale Car Dataset for Fine-Grained Categorization and Verification, In Computer Vision and Pattern Recognition (CVPR), 2015.