Downloads · 30 days
0
Mohith202/topics-in-ai-for-healthcare-assignment-2
topics-in-ai-for-healthcare-assignment-2 is a machine learning model from Mohith202. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Roll number 2025900010, random seed = 10. Companion weights for the assignment code (COVID-19 Radiography classification + lung segmentation + Grad-CAM, and Kvasir-SEG polyp detection).
Downloads · 30 days
0
Access
Public
Updated Sep 8, 2026
Repo size
254 MB
Likes
0
Public
Click a slice to open those files.
.pt254 MB · 100%
From the Hugging Face model README
Roll number 2025900010, random seed = 10. Companion weights for the assignment code (COVID-19 Radiography classification + lung segmentation + Grad-CAM, and Kvasir-SEG polyp detection).
| file | model | notes |
|---|---|---|
pretrained_xrv__rgb__aug-normal__init__main__seed10.pt | Model 1 - frozen TorchXRayVision DenseNet121 (densenet121-res224-all) + trainable 4-class head | test macro-F1 0.826, bal-acc 0.875, AUROC 0.974 |
student_densenet__rgb__aug-normal__init__main__seed10.pt | Model 2 - torchvision DenseNet121, ImageNet-initialized, fully fine-tuned | test macro-F1 0.967, bal-acc 0.960, AUROC 0.995 |
pretrained_xrv__lung_crop__aug-normal__init__ablation__seed10.pt | Model 1, lung-cropped input (seeded ablation) | test macro-F1 0.846 (+0.019 vs rgb) |
student_densenet__lung_crop__aug-normal__init__ablation__seed10.pt | Model 2, lung-cropped input (seeded ablation) | test macro-F1 0.936 (-0.031 vs rgb) |
unet_resnet34__main__seed10.pt | U-Net + ResNet-34 ImageNet encoder, lung segmentation | test Dice 0.989, IoU 0.979, Boundary-F1@2px 0.981, HD95 1.65 px |
Checkpoint dict: {"model_state", "config", "epoch", "val"/"val_dice", "classes"}.
| file | loss variant | notes |
|---|---|---|
kvasir_yolo/default_bce_seed10/weights/best.pt | stock YOLOv8 BCE classification loss | test [email protected] 0.904, [email protected]:0.95 0.700, P 0.900 / R 0.823 |
kvasir_yolo/focal_seed10/weights/best.pt | focal cls loss (gamma=2, patched into v8DetectionLoss) | test [email protected] 0.878, P 0.856 / R 0.835; more FPs, fewer FNs |
Code + frozen split files + full report are in the assignment submission. Splits
are created once with numpy.random.default_rng(10); seed 10 is used everywhere.
See RUN_COMMANDS.md in the submission for exact commands.