An Improved 3D Shape Recognition Method Based on Panoramic View

Table 2

Classification accuracy of various methods on the ModelNet-10 and ModelNet-40 datasets. Best results of this method are marked in bold font. The expanded training sets are expanded based on the balanced training set.

Method	ModelNet-10	ModelNet-40
Spherical harmonics descriptor [22]	79.79%	68.23%
Lightfield descriptor [23]	79.87%	75.47%
3D ShapeNets [7]	83.54%	77.32%
DeepPano [13]	88.66%	82.54%
Original training set	89.04%	80.68%
Balanced training set	89.80%	81.85%
Regularly expanded training set	87.54%	82.47%
Randomly expanded training set	85.62%	80.55%