Table 7 Ablation experiments on CA/MFC/BCEWITHLABELSMOOTHING modules. In this context, MFC refers to combining FPN and CSRA modules. The mean AUC scores are reported for 14 pathologies on the ChestX-Ray14 dataset.
From: Automated thorax disease diagnosis using multi-branch residual attention network
CA | MFC | BCEWithLabelSmoothing | Mean | FLOPS(G) | PARAMS(M) | GPU memory(G) | Inference time(s) |
|---|---|---|---|---|---|---|---|
✘ | ✘ | ✘ | 0.831 | 4.13 | 23.54 | 9.59 | 0.0268 |
✔ | ✘ | ✘ | 0.834 | 4.14 | 23.67 | 10.24 | 0.0281 |
✘ | ✔ | ✘ | 0.832 | 4.96 | 26.28 | 10.55 | 0.0273 |
✘ | ✘ | ✔ | 0.836 | 4.13 | 23.54 | 9.59 | 0.0268 |
✔ | ✔ | ✘ | 0.836 | 4.96 | 26.41 | 11.45 | 0.0289 |
✔ | ✔ | ✔ | 0.841 | 4.96 | 26.41 | 11.45 | 0.0289 |
