Add post-hoc analysis: confusion matrices, ROC, reliability, per-class F1, accuracy ranking

.gitattributes

@@ -33,3 +33,23 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_clip_openai.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_densenet121.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_dinov2_l.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_ensemble.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_inception_v3.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_resnet101.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_resnet50.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_retfound.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_swin_b.png filter=lfs diff=lfs merge=lfs -text
+analysis/confusion_matrices/cm_vgg19.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_clip_openai.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_densenet121.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_dinov2_l.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_ensemble.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_inception_v3.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_resnet101.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_resnet50.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_retfound.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_swin_b.png filter=lfs diff=lfs merge=lfs -text
+analysis/roc_curves/roc_vgg19.png filter=lfs diff=lfs merge=lfs -text

analysis/README.md

@@ -0,0 +1,13 @@
+# Analysis Artefacts
+
+Generated post-hoc from the test-set prediction JSONs (`results/*_test_preds.json`)
+and the ensemble report.
+
+## Contents
+- `confusion_matrices/cm_<model>.png` — per-class confusion matrix (counts overlaid on row-normalised heatmap)
+- `roc_curves/roc_<model>.png` — one-vs-rest ROC for every class + macro-AUC in title
+- `reliability_diagrams/reliability_<model>.png` — 10-bin reliability with ECE in title
+- `per_class/per_class_f1.png` — grouped bar chart, all 9 backbones side-by-side
+- `per_class/per_class_f1.json` — raw per-class F1 numbers
+- `accuracy_ranking.png` — horizontal bar chart of test accuracy
+- `ece_summary.json` — Expected Calibration Error per model

analysis/ece_summary.json

@@ -0,0 +1,12 @@
+{
+  "vgg19": 0.019883386493472457,
+  "resnet50": 0.03958680766850635,
+  "resnet101": 0.02942224809356465,
+  "densenet121": 0.017313514360019957,
+  "inception_v3": 0.026105817974059694,
+  "swin_b": 0.010641135098853942,
+  "clip_openai": 0.014958562176126485,
+  "dinov2_l": 0.03415616768925559,
+  "retfound": 0.026166619607560948,
+  "ensemble": 0.024558782145927764
+}

analysis/per_class/per_class_f1.json

@@ -0,0 +1,124 @@
+{
+  "classes": [
+    "Central Serous Chorioretinopathy",
+    "Diabetic Retinopathy",
+    "Disc Edema",
+    "Glaucoma",
+    "Healthy",
+    "Macular Scar",
+    "Myopia",
+    "Pterygium",
+    "Retinal Detachment",
+    "Retinitis Pigmentosa"
+  ],
+  "per_class_f1": {
+    "vgg19": [
+      0.9292035398230089,
+      0.9707927677329624,
+      0.9513888888888888,
+      0.7966101694915254,
+      0.8484304932735426,
+      0.8816568047337278,
+      0.8720271800679502,
+      1.0,
+      0.996078431372549,
+      0.9657320872274143
+    ],
+    "resnet50": [
+      0.918918918918919,
+      0.9674740484429065,
+      0.9477351916376306,
+      0.7986852917009039,
+      0.8574007220216606,
+      0.8807069219440353,
+      0.8537735849056604,
+      1.0,
+      0.996078431372549,
+      0.9657320872274143
+    ],
+    "resnet101": [
+      0.918918918918919,
+      0.9669421487603306,
+      0.9611307420494699,
+      0.8013411567476949,
+      0.8537906137184116,
+      0.8820960698689956,
+      0.859465737514518,
+      1.0,
+      0.996078431372549,
+      0.9657320872274143
+    ],
+    "densenet121": [
+      0.9140271493212669,
+      0.9715475364330326,
+      0.9513888888888888,
+      0.8158546655656482,
+      0.8616236162361623,
+      0.8833819241982507,
+      0.8662857142857143,
+      1.0,
+      0.996078431372549,
+      0.9657320872274143
+    ],
+    "inception_v3": [
+      0.918918918918919,
+      0.9689011748445059,
+      0.9513888888888888,
+      0.8045789043336059,
+      0.8534562211981567,
+      0.8875912408759125,
+      0.8504672897196262,
+      1.0,
+      0.996078431372549,
+      0.9657320872274143
+    ],
+    "swin_b": [
+      0.9074889867841409,
+      0.9667590027700831,
+      0.9513888888888888,
+      0.7436527436527437,
+      0.8237410071942446,
+      0.8831562974203339,
+      0.810304449648712,
+      1.0,
+      0.9921875,
+      0.9657320872274143
+    ],
+    "clip_openai": [
+      0.9203539823008849,
+      0.954829742876998,
+      0.9513888888888888,
+      0.7137614678899082,
+      0.8296943231441049,
+      0.8442503639010189,
+      0.8310010764262648,
+      1.0,
+      0.9883268482490273,
+      0.9657320872274143
+    ],
+    "dinov2_l": [
+      0.918918918918919,
+      0.9724896836313618,
+      0.9645390070921985,
+      0.8089700996677741,
+      0.8642413487133984,
+      0.8885630498533724,
+      0.8393285371702638,
+      1.0,
+      0.9921875,
+      0.9657320872274143
+    ],
+    "retfound": [
+      0.8595744680851064,
+      0.9548924358084664,
+      0.9134948096885813,
+      0.6984667802385008,
+      0.7932960893854749,
+      0.8096676737160121,
+      0.7918015102481122,
+      1.0,
+      0.9844961240310077,
+      0.9627329192546584
+    ]
+  }
+}