Abstract:
Existing vehicle re-identification methods use spatial pooling operations to aggregate feature maps extracted from off-the-shelf backbone networks like VGGNet, GoogLeNet, and ResNet. They neglect feature map spatial significance, lowering vehicle re-identification performance. First, an innovative spatial graph network (SGN) is proposed to elaborately explore feature map spatial significance in this paper. The SGN stacks SGs. SGs use spatial neighborhood relationships to determine edges between feature map nodes. SGN propagation aggregates each node and its spatial neighbors to the next SG. Each aggregated node is re-weighted with a learnable parameter on the next SG to determine significance. Second, a novel pyramidal graph network (PGN) is used to comprehensively explore feature maps’ spatial significance at multiple scales. The PGN pyramidally organizes SGNs to handle feature maps of different scales. After embedding the PGN behind a ResNet-50 backbone network, a hybrid pyramidal graph network (HPGN) is created. The proposed HPGN outperforms state-of-the-art vehicle re-identification methods in accuracy, parameter cost, and computation cost. Experiments show that the proposed PGN works for all backbone networks.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
