v1 fix

inference.py

@@ -54,8 +54,8 @@ class CompressionArtifactPredictor:
         self.quality_ranges = {
             'jpeg': (0, 100),
             'webp': (0, 100),
-            'avif': (0, 63),
-            'jxl': (0, 25)
+            'avif': (0, 100),
+            'jxl': (0, 100)
         }
 
     def predict(self, image: Image.Image) -> Dict[str, Dict[str, float]]:

model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05f4a29c7a0b12da38b7524eb32801b2bc05cd089101e017e2dee0b326347861
+oid sha256:1f260ddcbab4ecab0db436a2c72e146932bb10011b4ba841755fc63f49151979
 size 9386549

train.py

@@ -27,8 +27,8 @@ class Config:
     QUALITY_RANGES = {
         'jpeg': (0, 100),
         'webp': (0, 100),
-        'avif': (0, 63),
-        'jxl': (0, 25)
+        'avif': (0, 100),
+        'jxl': (0, 100)
     }
 
     # Training
@@ -57,9 +57,9 @@ def ensure_dir(path: str):
 def quality_to_normalized(quality: float, type: str) -> float:
     """Normalize JPEG quality [0,100] to [0,1]"""
     if type == "avif":
-        return quality / 63
+        return quality / 100
     if type == "jxl":
-        return quality / 25
+        return quality / 100
     return quality / 100.0
 
 
@@ -125,14 +125,14 @@ class CompressionDataset(Dataset):
             targets.append(quality_to_normalized(quality, "webp"))
             formats.append(Config.COMPRESSION_FORMATS.index("webp"))
 
-            quality = random.randint(0, 63)
+            quality = random.randint(0, 100)
             compressed = compress_image(image.copy(), "avif", quality)
             tensor = transforms.ToTensor()(compressed)
             images.append(tensor)
             targets.append(quality_to_normalized(quality, "avif"))
             formats.append(Config.COMPRESSION_FORMATS.index("avif"))
 
-            quality = random.randint(0, 25)
+            quality = random.randint(0, 100)
             compressed = compress_image(image.copy(), "jxl", quality)
             tensor = transforms.ToTensor()(compressed)
             images.append(tensor)