Fine-Grained Vehicle Classification with Unsupervised Parts Co-occurrence Learning

23 Jan 2019  ·  Sara Elkerdawy, Nilanjan Ray, Hong Zhang ·

Vehicle fine-grained classification is a challenging research problem with little attention in the field. In this paper, we propose a deep network architecture for vehicles fine-grained classification without the need of parts or 3D bounding boxes annotation. Co-occurrence layer (COOC) layer is exploited for unsupervised parts discovery. In addition, a two-step procedure with transfer learning and fine-tuning is utilized. This enables us to better fine-tune models with pre-trained weights on ImageNet in some layers while having random initialization in some others. Our model achieves 86.5% accuracy outperforming the state of the art methods in BoxCars116K by 4%. In addition, we achieve 95.5% and 93.19% on CompCars on both train-test splits, 70-30 and 50-50, outperforming the other methods by 4.5% and 8% respectively.

PDF Abstract


Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Fine-Grained Image Classification BoxCars116K ResNet152 + COOC Accuracy 86.57% # 1
Fine-Grained Image Classification CompCars Resnet50 + COOC Accuracy 95.6% # 3


No methods listed for this paper. Add relevant methods here