Upload export_medsiglip.py with huggingface_hub

export_medsiglip.py

@@ -239,7 +239,7 @@ def export_to_onnx(model, config: Config, quantize: bool = True):
     with torch.no_grad():
         traced = torch.jit.trace(wrapper, dummy_input)
 
-    # Export traced model to ONNX
+    # Export traced model to ONNX (fixed batch size for compatibility)
     torch.onnx.export(
         traced,
         dummy_input,
@@ -248,12 +248,8 @@ def export_to_onnx(model, config: Config, quantize: bool = True):
         opset_version=14,  # Use stable opset version
         do_constant_folding=True,
         input_names=['pixel_values'],
-        output_names=['seg_probs', 'plane_pred'],
-        dynamic_axes={
-            'pixel_values': {0: 'batch_size'},
-            'seg_probs': {0: 'batch_size'},
-            'plane_pred': {0: 'batch_size'}
-        }
+        output_names=['seg_probs', 'plane_pred']
+        # No dynamic_axes - use fixed batch size of 1 for mobile
     )
 
     # Verify ONNX model