Skip to content
TP
← Back to the lab
Archived2023

COVID-19 X-ray Detection

Published research pairing Xception with EfficientNet-B7 and ResNet152V2 across 11,000+ chest X-rays.

TensorFlowKerasSageMaker

The question

Does ensembling two strong backbones beat tuning one of them harder?

How it works

COLLECT11,302 images01AUGMENTbalance classes02TRAINdual backbones03ENSEMBLEweighted fusion04PUBLISHIJCRR05

Where the time went

effort
  • Data preparation48%
  • Model training32%
  • Write-up20%

How it progressed

relative momentum over the build

What I found

Yes, but the margin is small enough that data quality mattered more than either choice. Reached 94% accuracy.

Next experiment

AgentBridge

$./next-step

Most of these started as a question I could not answer by reading.

Send me a hard one.