Test run SRCNN+MSI on Truba

[nagellette]

Apply test runs on Truba to determine the relation between batch size, patch size, epoch size, GPU count, CPU count and model depth.

• [ ] Run models
• [ ] Evaluate results
• [ ] Create speed up efficiency graphs (? - metioned in the TIK on June)

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 512x512 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 1 # Cores: 10 Initial raw usage in seconds: 39644 Final raw usage in seconds: 85184

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 512x512 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 2 # Cores: 10 Initial raw usage in seconds: 85184 Final raw usage in seconds: 133344

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 1024x1024 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 1 # Cores: 10 Initial raw usage in seconds: 133344 Final raw usage in seconds: 292374

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 1024x1024 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 2 # Cores: 10 Initial raw usage in seconds: 292374 Final raw usage in seconds: 452974

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 4 Batch count: 512x512 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 4 # Cores: 10 Initial raw usage in seconds: 455314 Final raw usage in seconds: 537754

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 4 Batch count: 512x512 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 4 # Cores: 10 Initial raw usage in seconds: 650340 Final raw usage in seconds: 733020

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands + 3 GPS) Batch size: 4 Batch count: 512x512 # epochs: 10 Epoch size: 1000 Cluster: Akya-cuda # GPU: 4 # Cores: 10 Initial raw usage in seconds: 733020 Final raw usage in seconds: 839800

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 512x512 # epochs: 10 Epoch size: 2000 Cluster: Akya-cuda # GPU: 1 # Cores: 10 Initial raw usage in seconds: 839800 Final raw usage in seconds: 927380

[nagellette]

Model: UNET Dataset: Montreal-small (4 MSI bands) Batch size: 2 Batch count: 1024x1024 # epochs: 10 Epoch size: 2000 Cluster: Akya-cuda # GPU: 1 # Cores: 10 Initial raw usage in seconds: 1108100 Final raw usage in seconds: 1431260

[nagellette]

Model: SRCNN+UNET Dataset: Montreal-small (4 MSI bands) Batch size: 1 Batch count: 1024x1024 # epochs: 40 Epoch size: 2000 Cluster: Akya-cuda # GPU: 4 # Cores: 10 Initial raw usage in seconds: 1433434 Final raw usage in seconds: N/A