Research Article
Deep-Sea: A Reconfigurable Accelerator for Classic CNN
Table 5
Performance statistics of YOLOv3-tiny each layer under 4-16 architecture.
| | Input FMP | Output FMP | CONV size | Ops | Processing time | Performance (GOP/s) |
| | | 3 | 74760192 | 1.6 ms | 93.45 | | | 3 | 199360512 | 2.29 ms | 174.11 | | | 3 | 199360512 | 2.36 ms | 168.95 | | | 3 | 199360512 | 2.01 ms | 198.37 | | | 3 | 199360512 | 2.43 ms | 164.08 | | | 3 | 199360512 | 4.07 ms | 97.97 | | | 3 | 797442048 | 16.28 ms | 97.97 | | | 1 | 44302336 | 4.67 ms | 18.97 | | | 3 | 199360512 | 4.07 ms | 97.97 | | | 1 | 22151168 | 2.34 ms | 18.93 | | | 1 | 5537792 | 0.58 ms | 19.10 | | | 3 | 149520384 | 3.05 ms | 98.05 | | | 1 | 11075584 | 1.2 ms | 18.46 | | Total | | | 2300980176 | 46.95 | 98.01 |
|
|
