ultralytics 8.2.84 new SAM flexible imgsz inference (#15882)

Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

ultralytics/models/sam/modules/tiny_encoder.py

@@ -982,10 +982,31 @@ class TinyViT(nn.Module):
             layer = self.layers[i]
             x = layer(x)
         batch, _, channel = x.shape
-        x = x.view(batch, 64, 64, channel)
+        x = x.view(batch, self.patches_resolution[0] // 4, self.patches_resolution[1] // 4, channel)
         x = x.permute(0, 3, 1, 2)
         return self.neck(x)
 
     def forward(self, x):
         """Performs the forward pass through the TinyViT model, extracting features from the input image."""
         return self.forward_features(x)
+
+    def set_imgsz(self, imgsz=[1024, 1024]):
+        """
+        Set image size to make model compatible with different image sizes.
+
+        Args:
+            imgsz (Tuple[int, int]): The size of the input image.
+        """
+        imgsz = [s // 4 for s in imgsz]
+        self.patches_resolution = imgsz
+        for i, layer in enumerate(self.layers):
+            input_resolution = (
+                imgsz[0] // (2 ** (i - 1 if i == 3 else i)),
+                imgsz[1] // (2 ** (i - 1 if i == 3 else i)),
+            )
+            layer.input_resolution = input_resolution
+            if layer.downsample is not None:
+                layer.downsample.input_resolution = input_resolution
+            if isinstance(layer, BasicLayer):
+                for b in layer.blocks:
+                    b.input_resolution = input_resolution