input1 --> label1 — 32% data
input2 --> label2 — 32% data
input3 --> label3 — 32% data
input4 --> label1,label2 — 1% data
input5 --> label1,label3 — 1% data
input6 --> label2,label3 — 1% data
input7 --> label1,label2,label3 — 1% data
解决思路：
《Learning a Deep ConvNet for Multi-label Classification with Partial Labels》
https://arxiv.org/pdf/1902.09720.pdf
