# 2nd Place Solution

Solution Summary
My solution is based on a two-model ensemble combining EVA-X and CheXFound, each trained separately for Division A and Division B.
The pipeline focuses on optimized preprocessing, strong augmentations, checkpoint averaging, and EMA to achieve stable and high-performing predictions.
Approach
I used a 600×600 preprocessing pipeline for all images to reduce I/O overhead and improve training speed.
Then, I trained two independent high-performance vision models:
EVA-X (448×448)
6 training epochs
Layer-wise learning rates
Strong geometric + photometric augmentations
Focal Loss + BCE
Mixed Precision (AMP)
Exponential Moving Average (EMA)
Checkpoint averaging (epochs 5 + 6)
CheXFound (512×512)
6 epochs (3 + 3)
Only the GLoRIA head fine-tuned
Backbone frozen
Same augmentations and optimizers as EVA-X
The final prediction is a simple average of:
EVA-X (epoch 5 + TTA), 
EVA-X (epoch 6 + TTA), 
CheXFound (epoch 5), 
CheXFound (epoch 6)
This ensemble produced the best and most stable leaderboard performance in both divisions.
For the full explanation, preprocessing, training notebooks, inference code, and model configurations, please see the complete GitHub repository:
👉 GitHub Repository:
[https://github.com/Masry5/Grand-X-Ray-Slam-competition](https://github.com/Masry5/Grand-X-Ray-Slam-competition)