feat: diverse examples showcasing 5 capabilities + real images

.gitattributes

@@ -38,3 +38,9 @@ examples/bottle_broken_small.png filter=lfs diff=lfs merge=lfs -text
 examples/bottle_contamination.png filter=lfs diff=lfs merge=lfs -text
 examples/bottle_good.png filter=lfs diff=lfs merge=lfs -text
 examples/bottle_test_good.png filter=lfs diff=lfs merge=lfs -text
+examples/cable_cut.png filter=lfs diff=lfs merge=lfs -text
+examples/cable_good.png filter=lfs diff=lfs merge=lfs -text
+examples/capsule_crack.png filter=lfs diff=lfs merge=lfs -text
+examples/capsule_good.png filter=lfs diff=lfs merge=lfs -text
+examples/metal_nut_good.png filter=lfs diff=lfs merge=lfs -text
+examples/metal_nut_scratch.png filter=lfs diff=lfs merge=lfs -text

app.py

@@ -102,34 +102,39 @@ Upload a **reference (good) image** and a **test image**, then ask the model to
 """
 
 EXAMPLES = [
+    # 1. Anomaly Discrimination (yes/no) — bottle
     [
         "examples/bottle_good.png",
         "examples/bottle_contamination.png",
-        "The first image is good. Is there any anomaly in the second image? A.yes, B.no. Please answer the letter only.",
+        "The first image is a normal sample. Is there any anomaly in the second image? A. Yes B. No. Please answer the letter only.",
         128,
     ],
+    # 2. Defect Classification — metal nut
     [
-        "examples/bottle_good.png",
-        "examples/bottle_broken_large.png",
-        "The first image is good. Is there any anomaly in the second image? A.yes, B.no. Please answer the letter only.",
+        "examples/metal_nut_good.png",
+        "examples/metal_nut_scratch.png",
+        "The first image is a normal sample. What is the type of defect in the second image? A. Bent B. Scratch C. Color D. No defect. Please answer the letter only.",
         128,
     ],
+    # 3. Defect Description — cable
     [
-        "examples/bottle_good.png",
-        "examples/bottle_test_good.png",
-        "The first image is good. Is there any anomaly in the second image? A.yes, B.no. Please answer the letter only.",
-        128,
+        "examples/cable_good.png",
+        "examples/cable_cut.png",
+        "The first image is a normal sample. Compared with the first image, please describe the anomaly in the second image in detail.",
+        256,
     ],
+    # 4. Defect Localization — capsule
     [
-        "examples/bottle_good.png",
-        "examples/bottle_contamination.png",
-        "The first image is a good sample. Compared with the first image, please describe the anomaly in the second image in detail.",
+        "examples/capsule_good.png",
+        "examples/capsule_crack.png",
+        "The first image is a normal sample. Please locate the defects within the second image with bounding box in JSON format.",
         256,
     ],
+    # 5. Object Analysis — bottle (good vs good)
     [
         "examples/bottle_good.png",
-        "examples/bottle_broken_small.png",
-        "The first image is a good sample. Compared with the first image, please describe the anomaly in the second image in detail.",
+        "examples/bottle_test_good.png",
+        "The first image is a normal sample. Please describe the object in the second image and analyze whether it has any quality issues.",
         256,
     ],
 ]