update

app.py

@@ -282,7 +282,7 @@ with gr.Blocks() as demo:
             # 添加fp16和try_run复选框
             fp16 = gr.Checkbox(label="FP16", value=False)
             onnxsim = gr.Checkbox(label="ONNX export simplify model", value=False)
-            opset = gr.Number(label="ONNX export opset version", value=11, precision=0)
+            opset = gr.Number(label="ONNX export opset version, suggest 9/11/13/16/17/18", value=13, precision=0)
             try_run = gr.Checkbox(label="MNNSR test", value=False)
             convert_btn = gr.Button("Run")
         with gr.Column():
@@ -329,19 +329,18 @@ with gr.Blocks() as demo:
                 )
 
             if not model_input:
-                log += f'\n模型文件下载失败\n'
+                log += '模型文件下载失败\n'
                 print_log(task_counter, f'模型文件载', '失败')
                 yield None, None, log, None
                 return
             
-            log += f'\n模型文件已下载到: {model_input}\n'
+            log += f'模型文件已下载到: {model_input}\n'
             print_log(task_counter, f'模型文件已下载到: {model_input}', '完成')
             yield None, None, log, None
         elif input_type == model_type_opt[1] and file_input:
             model_input = file_input
         else:
-            # 改为通过yield返回错误日志
-            log = '\n请选择输入类型并提供有效的输入！'
+            log = '请选择输入类型并提供有效的输入！\n'
             yield None, None, log,None
             return
         
@@ -361,10 +360,10 @@ with gr.Blocks() as demo:
             if mnn_path: 
                 if try_run:
                     print_log(task_counter, f'测试模型: {mnn_path}', '开始')
-                    processed_image_np = modelTest_for_gradio(mnn_path, "./sample.jpg", int(tilesize), 0)
+                    processed_image_np, load_time, infer_time = modelTest_for_gradio(mnn_path, "./sample.jpg", int(tilesize), 0)
                     processed_image_pil = Image.fromarray(cv2.cvtColor(processed_image_np, cv2.COLOR_BGR2RGB))
                     # processed_image_pil = Image.fromarray(processed_image_np)
-                    yield onnx_path, mnn_path, log+process_log, processed_image_pil
+                    yield onnx_path, mnn_path, log+process_log+f"MNNSR 加载模型用时 {load_time:.4f} 秒, 推理({tilesize} px)用时 {infer_time:.4f} 秒", processed_image_pil
                 else:
                     yield onnx_path, mnn_path, log+process_log, None
                 return
@@ -373,7 +372,7 @@ with gr.Blocks() as demo:
         process_model,
         inputs=[input_type, url_input, file_input, tilesize, fp16, onnxsim, opset, try_run],
         outputs=[onnx_output, mnn_output, log_box, img_output],
-        api_name="convert_model"
+        api_name="convert_nmm_model"
     )
 
     # 将示例移至底部并包裹在列组件中

app_mnnsr.py

@@ -7,10 +7,8 @@ from PIL import Image
 from mnnsr import modelTest_for_gradio
 
 def gradio_interface(modelPath, input_image):
-    processed_image_np = modelTest_for_gradio(modelPath, input_image)
-    
+    processed_image_np, load_time, infer_time = modelTest_for_gradio(modelPath, input_image)
     processed_image_pil = Image.fromarray(cv2.cvtColor(processed_image_np, cv2.COLOR_BGR2RGB))
-    
     return processed_image_pil
 
 # 创建Gradio界面

mnnsr.py

@@ -4,6 +4,7 @@ import MNN
 import numpy as np
 import cv2
 from PIL import Image
+import time
 
 
 # 复制原始modelTest函数中的必要函数
@@ -64,11 +65,14 @@ def modelTest_for_gradio(modelPath, image_path, tilesize = 0, backend = 3):
     model_name = os.path.basename(modelPath)
     if "-Grayscale" in model_name:
         model_channel = 1
-    elif "-4ch" in model_name:
+    elif "-4ch" in model_name or "RGBA" in model_name:
         model_channel = 4
     else:
         model_channel = 3
 
+    # 记录模型加载开始时间
+    load_start_time = time.time()
+    # 加载模型（计时范围内）
     net = MNN.Interpreter(modelPath)
     # set 9 for Session_Backend_Auto, Let BackGround Tuning
     net.setSessionMode(9)
@@ -79,8 +83,14 @@ def modelTest_for_gradio(modelPath, image_path, tilesize = 0, backend = 3):
     config['backend'] = backend
     #config['precision'] = "low"
     session = net.createSession(config)
-    
     print("Run on backendtype: %d \n" % net.getSessionInfo(session, 2))
+    inputTensor = net.getSessionInput(session)
+    net.resizeTensor(inputTensor, (1, model_channel, tilesize, tilesize))
+    net.resizeSession(session)
+    # 计算模型加载耗时
+    load_time = time.time() - load_start_time
+    print(f"Load mnn model: {load_time:.4f} sec")
+    
     # 读取图像
     image = cv2.imread(image_path)
     if image.ndim == 2:
@@ -93,6 +103,9 @@ def modelTest_for_gradio(modelPath, image_path, tilesize = 0, backend = 3):
     
     image = cv2.resize(image, (tilesize, tilesize))
 
+    # 记录推理开始时间
+    infer_start_time = time.time()
+
     # 处理通道数不匹配的情况
     if image_channel == 3:
         if model_channel == 1:
@@ -119,23 +132,17 @@ def modelTest_for_gradio(modelPath, image_path, tilesize = 0, backend = 3):
     # display(Image(data=cv2.imencode('.jpg', image)[1].tobytes()))
     # print("image.shape=", image.shape)
     image = image/255.0
-    #preprocess it
     if model_channel>=3:
         image = image.transpose((2, 0, 1))
-    #change numpy data type as np.float32 to match tensor's format
     image = image.astype(np.float32)
-    #cv2 read shape is NHWC, Tensor's need is NCHW,transpose it
+
     tmp_input = MNN.Tensor((1, model_channel, tilesize, tilesize), MNN.Halide_Type_Float, image, MNN.Tensor_DimensionType_Caffe)
-    
-    # input
-    inputTensor = net.getSessionInput(session)
-    net.resizeTensor(inputTensor, (1, model_channel, tilesize, tilesize))
-    net.resizeSession(session)
-    inputTensor.copyFrom(tmp_input)
-    # infer
+    inputTensor.copyFrom(tmp_input)   
+
+    # 执行推理
     net.runSession(session)
+    
     outputTensor = net.getSessionOutput(session)
-    # output
     outputShape = outputTensor.getShape()
     print("outputShape",outputShape)
     outputHost = createTensor(outputTensor)
@@ -143,8 +150,11 @@ def modelTest_for_gradio(modelPath, image_path, tilesize = 0, backend = 3):
     
     outimage = process_image(outputHost.getData(), outputShape[2], outputShape[3], outputShape[1], color='RGB')
     
-    # 返回处理后的图像(numpy数组)
-    return outimage
+    # 计算推理耗时
+    infer_time = time.time() - infer_start_time
+    print(f"Infer latency time: {infer_time:.4f} sec")
+    # 返回图像、加载时间、推理时间
+    return outimage, load_time, infer_time
 
 # def gradio_interface(modelPath, input_image):
 #     processed_image_np = modelTest_for_gradio(modelPath, input_image)

pth2onnx.py

@@ -70,7 +70,7 @@ def convert_pth_to_onnx(pth_path: str, onnx_path: str=None, channel:int=0, tiles
         filename = os.path.basename(pth_path).upper()
         pattern = f'(^|[_-])({scale}X|X{scale})([_-]|$)'
         if re.search(pattern, filename):
-            print(f'文件名 {filename} 包含倍率信息。')
+            print(f'File name contains scale info: {filename} ')
         else:
             base_path = f"{base_path}-x{scale}"
         # print("final use_fp16", str(use_fp16) )
@@ -81,8 +81,9 @@ def convert_pth_to_onnx(pth_path: str, onnx_path: str=None, channel:int=0, tiles
     elif output_folder:
         onnx_path = os.path.join(output_folder, onnx_path)
         
-    print(f"output_folder: {output_folder}, onnx_path: {onnx_path}")
+    # print(f"output_folder: {output_folder}, onnx_path: {onnx_path}")
 
+    print(f"ONNX model exporting...")
     try:
         # Export the model
         torch.onnx.export(
@@ -99,7 +100,7 @@ def convert_pth_to_onnx(pth_path: str, onnx_path: str=None, channel:int=0, tiles
                 "output": {0: "batch_size", 2: "height", 3: "width"},# Batch, H, W can vary
             }
         )
-        print(f"ONNX export successful: {onnx_path}")
+        print(f"ONNX model export successful: {onnx_path}")
 
         # Optional: Simplify the ONNX model
         if simplify_model:
@@ -115,11 +116,11 @@ def convert_pth_to_onnx(pth_path: str, onnx_path: str=None, channel:int=0, tiles
                 return onnx_path
             
             os.remove(onnx_path)
-            print(f"文件大小不足 {min_size} 字节，已删除无效文件")
+            print(f"ONNX model has unexpected file size ({file_size} bytes), deleted invalid file")
         return ""
 
     except Exception as e:
-        print(f"导出失败: {e}")
+        print(f"ONNX model export error: {e}")
         return ""
 
 if __name__ == "__main__":