ok

Author: johndpope

.gitignore

@@ -11,3 +11,4 @@ __pycache__/vit.cpython-311.pyc
 __pycache__/helper.cpython-311.pyc
 actions-runner/*
 imf_encoder.onnx
+imf_encoder_web.onnx

onnxconv.py

@@ -4,7 +4,8 @@
 import torch.nn as nn
 from PIL import Image
 from torchvision import transforms
-
+from onnxconverter_common import float16
+import onnx
 
 class IMFDecoder(nn.Module):
     def __init__(self, model):
@@ -63,7 +64,22 @@ def print_model_structure(model):
         if hasattr(module, 'bias') and module.bias is not None:
             print(f"  Bias shape: {module.bias.shape}")
 
-# Adjusted export_to_onnx function
+
+def convert_int64_to_int32(model):
+    for tensor in model.graph.initializer:
+        if tensor.data_type == onnx.TensorProto.INT64:
+            tensor.data_type = onnx.TensorProto.INT32
+            tensor.int64_data = tensor.int64_data.astype(np.int32)
+
+    for node in model.graph.node:
+        for attr in node.attribute:
+            if attr.type == onnx.AttributeProto.INT:
+                attr.i = int(attr.i)
+            elif attr.type == onnx.AttributeProto.INTS:
+                attr.ints[:] = [int(i) for i in attr.ints]
+
+    return model
+
 def export_to_onnx(model, x_current, x_reference, file_name):
     try:
         print("Model structure before tracing:")
@@ -97,7 +113,7 @@ def export_to_onnx(model, x_current, x_reference, file_name):
             opset_version=11,
             do_constant_folding=True,
             input_names=['x_current', 'x_reference'],
-            output_names=['f_r', 't_r', 't_c'],  # Adjusted output names
+            output_names=['f_r', 't_r', 't_c'],
             dynamic_axes={
                 'x_current': {0: 'batch_size'},
                 'x_reference': {0: 'batch_size'},
@@ -108,6 +124,23 @@ def export_to_onnx(model, x_current, x_reference, file_name):
             verbose=True
         )
         print(f"Model exported successfully to {file_name}")
+
+        # Load the ONNX model
+        onnx_model = onnx.load(file_name)
+
+        # Convert int64 to int32
+        print("Converting int64 to int32...")
+        onnx_model = convert_int64_to_int32(onnx_model)
+
+        # Optionally, convert float32 to float16 to reduce model size
+        print("Converting float32 to float16...")
+        onnx_model = float16.convert_float_to_float16(onnx_model)
+
+        # Save the converted model
+        web_compatible_file = file_name.replace('.onnx', '_web.onnx')
+        onnx.save(onnx_model, web_compatible_file)
+        print(f"Web-compatible model saved as {web_compatible_file}")
+
     except Exception as e:
         print(f"Error during ONNX export: {str(e)}")
         import traceback
@@ -128,9 +161,9 @@ def export_to_onnx(model, x_current, x_reference, file_name):
 # Load images and preprocess
 def load_image(image_path):
     transform = transforms.Compose([
-        transforms.Resize((256, 256)),  # Adjust as per your model's requirements
+        transforms.Resize((256, 256)),
         transforms.ToTensor(),
-        transforms.Normalize(mean=[0.485, 0.456, 0.406],  # Adjust as per your model's requirements
+        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                              std=[0.229, 0.224, 0.225])
     ])
     image = Image.open(image_path).convert('RGB')
@@ -141,4 +174,4 @@ def load_image(image_path):
 x_reference = load_image("x_reference.png")
 
 # Export the model
-export_to_onnx(encoder_model, x_current, x_reference, "imf_encoder.onnx")
+export_to_onnx(encoder_model, x_current, x_reference, "imf_encoder.onnx")

requirements.txt

@@ -15,4 +15,7 @@ onnxruntime
 opencv-python
 pymatting
 decord
-mediapipe
+mediapipe
+
+onnx 
+onnxconverter_common