This repository proposes VQ-BNN inference, a temporal smoothing of BNN's recent predictions, in order to improve the computational performance of BNN. In addition, it also proposes VQ-DNN, which is a temporal smoothing of deterministic NN's predictions. Experimental results show that the computational performance of VQ-BNN is almost the same as that of deterministic NN (DNN), and the predictive performance is comparable to or even superior to that of BNN. Similarly, the predictive performance of VQ-DNN is better than that of DNN.
This material provides predictive results and predictive uncertainties of DNN, VQ-DNN, BNN, and VQ-BNN on five different sequences. According to these qualitative results, the predictive results of DNN and BNN are noisy. Their classification results for an object change irregularly and randomly. In contrast, the predictive results of VQ-DNN and VQ-BNN are stabilized. Their predictive results change smoothly.
| Input | DNN (11 FPS) |
VQ-DNN (10 FPS) |
BNN (0.8 FPS) |
VQ-BNN (9 FPS) |
|---|---|---|---|---|
| Seq 1 | ||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| Seq 2 | ||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| Seq 3 | ||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| Seq 4 | ||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| Seq 5 | ||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
|