Table 2.
Description of the Deep Learning Models and Training Approaches at UCSD and UTokyo
| Parameter | UCSD Deep Learning Model | UTokyo Deep Learning Models |
|---|---|---|
| Network architecture | ResNet5030 | ResNet3430 |
| Weight initialization | Pretraining on ImageNet24 | Pretraining on ImageNet24 |
| Data augmentation | Translation, horizontal flipping | Translation, scaling, rotation, and horizontal flipping |
| Training datasets | ||
| Original | DIGS/ADAGES, (ImageNet pretraining) | MRCH, (ImageNet pretraining) |
| Sequential | DIGS/ADAGES, weights updated using MRCH, (ImageNet pretraining) | MRCH, weights updated using DIGS/ADAGES, (ImageNet pretraining) |
| Combined | DIGS/ADAGES & MRCH (ImageNet pretraining) | MRCH & DIGS/ADAGES (ImageNet pretraining) |