ct-transformer

Swin-based chest X-ray multi-label classifier exported from CT-transformer.

Model

• Backbone: swin_base_patch4_window7_224
• Input size: 512x512
• Labels: 15
• Best validation macro AUC-ROC: 0.9664756266963344

Labels

[
  "Aortic enlargement",
  "Atelectasis",
  "Calcification",
  "Cardiomegaly",
  "Consolidation",
  "ILD",
  "Infiltration",
  "Lung Opacity",
  "Nodule/Mass",
  "Other lesion",
  "Pleural effusion",
  "Pleural thickening",
  "Pneumothorax",
  "Pulmonary fibrosis",
  "No finding"
]

Colab Usage

!pip install torch torchvision timm safetensors huggingface_hub pillow

from huggingface_hub import snapshot_download
from PIL import Image
import torch

repo_dir = snapshot_download("sbandred/ct-transformer")

import sys
sys.path.append(repo_dir)
from modeling_ct_transformer import load_model

model, config = load_model(repo_dir, device="cuda" if torch.cuda.is_available() else "cpu")

image = Image.open("/content/example.png").convert("RGB")

This bundle intentionally exports inference weights only, not optimizer or scheduler state.