Spaces:
Sleeping
Sleeping
File size: 1,091 Bytes
8960670 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 | #!/usr/bin/env python3
"""
Baseline 2D ResNet-18 (Slice-Level) β Standalone module.
This module provides the ResNet2D18SliceLevel class as a standalone import
per the ablation study specification. The actual implementation lives in
baselines.py; this file re-exports it for clean per-model imports.
Usage:
from src.models.baseline_resnet2d import ResNet2D18SliceLevel
model = ResNet2D18SliceLevel(num_classes=1)
Architecture:
- Backbone: torchvision resnet18 (2D)
- Input conv modified: 3-channel β 1-channel (grayscale CT slices)
- Each 3D volume (B, 1, D, H, W) is processed slice-by-slice:
1. Reshape β (B*D, 1, H, W)
2. Forward through ResNet-18 backbone β (B*D, 512)
3. Reshape back β (B, D, 512)
4. Global Average Pool over depth β (B, 512)
5. Classification head β (B, 1)
- Forward signature: forward(nodule_patch, context_patch=None)
(context_patch is ignored β 2D baseline uses nodule slices only)
"""
from src.models.baselines import ResNet2D18SliceLevel
__all__ = ['ResNet2D18SliceLevel']
|