Table 4 Performance of FCN without quantization (base case) and with direct and retrained quantizations. The table shows the performance of FCN in terms of percentage accuracy (PA, mean intersection over union (mean IoU), and MA). Notice the severe degradation in performance with direct quantization of the network, while the performance is almost retained by the retrained network in the first four configurations. A retrained network with configuration E77-D77-SC0 produces the best performance.
From: Optimizing FCN for devices with limited resources using quantization and sparsity enhancement
FCN without quantization | FCN with quantized networks | ||||||||
|---|---|---|---|---|---|---|---|---|---|
Percentage accuracy (%) | Config # | Percentage accuracy with direct quantization (%) | Percentage accuracy with retrained quantization (%) | ||||||
PA | Mean IOU | PA | PA | Mean IOU | MA | PA | Mean IOU | MA | |
91 | 62 | 78 | E33-D33-SC0 | 75 | 4 | 5 | 83 | 46 | 58 |
E77-D77-SC0 | 76 | 7 | 8 | 90 | 59 | 77 | |||
E73-D33-SC0 | 75 | 4 | 5 | 88 | 53 | 74 | |||
E37-D33-SC0 | 75 | 4 | 5 | 89 | 56 | 75 | |||
E33-D33-SC3 | 75 | 4 | 5 | 75 | 4 | 10 | |||
E77-D77-SC7 | 76 | 7 | 8 | 76 | 4 | 5 | |||
