Two networks have been created, which can be found in: models/networks/classification/
- VGG16
- MiniNet: Implemented network that stands for a convolutional neural network with low number of parameters.
| Task A - Run Code | Task B - Train with KITTI | Task C - New Architechtures | Task E - Documentation |
|---|---|---|---|
| Analyse Datasets | FineTune | Implement MiniNet | Report |
| Fine-Tune VGG16 for TT100K and BelgiumTSC | Train from Scratch | Slides |
The following table shows the results obtained from the different tasks. We have used two networks, the well known VGG16 and our own CNN, MiniNet. The experiments have been done using the following datasets: TT100k, BelgiumTSC and KITTI.
| Network | Experiment | TT100K | Belgium | KITTI | |||
|---|---|---|---|---|---|---|---|
| Val | Test | Val | Test | Val | Test | ||
| VGG16 | Basic(ImageNet) | 89,32 | 96.06 | 96.22 | 95.22 | 98.37 | - |
| VGG16 | FineTune with TT100K | - | - | 96.39 | 96.39 | 97.84 | - |
| VGG16 | From scratch | - | - | - | - | 97.30 | - |
| MiniNet | From scratch | 84.46 | 92.32 | 90.27 | 90.28 | 92.48 | - |