<< All Blogs
Week #8
20 Dec 2018
Done:
- Experiments with KITTI semantic segmentation datasets. It consists of 200 semantically annotated train as well as 200 test images corresponding to the KITTI Stereo and Flow Benchmark 2015.
- Github link Completed:
- Dataloader constructed
- Training the data
- Note: 1 batch, 1 image consume about 3000 Mbps memory in GPU
- Experiments:
| Number |
Data source |
Training Sample |
epoch of training |
Batch size |
learning rate |
momentum |
Mimimum loss attained |
| 01 |
KITTI |
#200 |
2 |
2 |
0.001 |
0.9 |
0.798 |
| 02 |
KITTI |
#200 |
10 |
2 |
0.001 |
0.9 |
0.603 |
| 03 |
KITTI |
#160 |
1 |
1 |
0.25 |
0.9 |
0.552 |
| 04 |
KITTI |
#160 |
1 |
1 |
0.01 |
0.9 |
0.577 |
To do:
- Tune the Hyper-parameter.
- Go back to tutorial at cs231n for Hyperparameter tuning.