Research Article

TANet: A Tiny Plankton Classification Network for Mobile Devices

Table 1

Design architectures and evaluation on plankton dataset. Our experiments are based on the caffe [32] framework, and the model size indicates the size of the hard disk space actually occupied by the caffe model. Inference time is measured on the NVIDIA GeForce Titan X Pascal 12 GB.

NetworkTop-1 (%)Top-5 (%)Inference time (ms)Model size (kB)

Figure 4(a)75.995.621.6642
Figure 4(b)76.195.927.3642
Figure 4(c) (TANet)76.596.331.8648