unet_semantic_segmentation
Posted by ttmax
Copy the project to Neural Network Console Cloud
Simple example of semantic segmentation with UNet like architecture.
U-Net: Convolutional Networks for Biomedical Image Segmentation
Olaf Ronneberger, Philipp Fischer, Thomas Brox
https://arxiv.org/abs/1505.04597
dataset-require=PASCALVOC2012_Segmentation
Network Architecture : UNet
| Type | Value |
|---|---|
| Output | 65,536,000 |
| CostParameter | 33,614,336 |
| CostAdd | 36,241,408 |
| CostMultiply | 36,175,872 |
| CostMultiplyAdd | 84,439,728,128 |
| CostDivision | 0 |
| CostExp | 0 |
| CostIf | 36,175,872 |
Network Architecture : DSUnit
| Type | Value |
|---|---|
| Output | 25,952,256 |
| CostParameter | 39,104 |
| CostAdd | 8,388,608 |
| CostMultiply | 8,388,608 |
| CostMultiplyAdd | 2,529,165,312 |
| CostDivision | 0 |
| CostExp | 0 |
| CostIf | 8,388,608 |
Network Architecture : USUnit
| Type | Value |
|---|---|
| Output | 113,246,208 |
| CostParameter | 206,336 |
| CostAdd | 33,554,432 |
| CostMultiply | 33,554,432 |
| CostMultiplyAdd | 53,955,526,656 |
| CostDivision | 0 |
| CostExp | 0 |
| CostIf | 33,554,432 |
Network Architecture : Train
| Type | Value |
|---|---|
| Output | 2,091,044 |
| CostParameter | 33,614,336 |
| CostAdd | 36,503,552 |
| CostMultiply | 36,438,016 |
| CostMultiplyAdd | 84,439,728,128 |
| CostDivision | 65,536 |
| CostExp | 65,536 |
| CostIf | 36,306,944 |
Network Architecture : Validation
| Type | Value |
|---|---|
| Output | 1,239,076 |
| CostParameter | 33,614,336 |
| CostAdd | 36,438,016 |
| CostMultiply | 36,438,016 |
| CostMultiplyAdd | 84,439,728,128 |
| CostDivision | 0 |
| CostExp | 0 |
| CostIf | 36,306,944 |
Network Architecture : Runtime
| Type | Value |
|---|---|
| Output | 1,551,488 |
| CostParameter | 33,614,336 |
| CostAdd | 36,630,040 |
| CostMultiply | 36,564,504 |
| CostMultiplyAdd | 84,439,728,128 |
| CostDivision | 64,009 |
| CostExp | 64,009 |
| CostIf | 36,495,912 |
Training Procedure : Optimizer
Optimize network “Train” using “Training” dataset.
- Batch size : 1
- Solver : Momentum
- Learning rate: 0.002
- Momentum : 0.99
- Weight decay : 0.0005
References
- Sony Corporation. Neural Network Console : Not just train and evaluate. You can design neural networks with fast and intuitive GUI. https://dl.sony.com/
- Sony Corporation. Neural Network Libraries : An open source software to make research, development and implementation of neural network more efficient. https://nnabla.org/
- Convolution – Chen et al., DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. https://arxiv.org/abs/1606.00915, Yu et al., Multi-Scale Context Aggregation by Dilated Convolutions. https://arxiv.org/abs/1511.07122
- BatchNormalization – Ioffe and Szegedy, Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. https://arxiv.org/abs/1502.03167
- ReLU – Vinod Nair, Geoffrey E. Hinton. Rectified Linear Units Improve Restricted Boltzmann Machines. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.165.6419&rep=rep1&type=pdf
- Momentum – Ning Qian : On the Momentum Term in Gradient Descent Learning Algorithms. http://www.columbia.edu/~nq6/publications/momentum.pdf