add ax650 backend

README.md

@@ -1 +1,51 @@
-# CLIP-ONNX
+# CLIP
+
+## ONNX
+
+```
+./main -e ../onnx_models/CLIP_encoder.onnx -d ../onnx_models/decode.onnx -v ../vocab.txt -i ../images/ -t ../text.txt 
+
+inputs: 
+             input.1: 1 x 3 x 224 x 224
+output: 
+                2270: 1 x 512
+
+per image:
+                 image path\text|                            bird|                             cat|                             dog|
+              ../images/bird.jpg|                            1.00|                            0.00|                            0.00|
+               ../images/cat.jpg|                            0.00|                            0.99|                            0.01|
+         ../images/dog-chai.jpeg|                            0.00|                            0.02|                            0.98|
+
+
+per text:
+                 text\image path|              ../images/bird.jpg|               ../images/cat.jpg|         ../images/dog-chai.jpeg|
+                            bird|                            0.96|                            0.01|                            0.03|
+                             cat|                            0.00|                            0.91|                            0.09|
+                             dog|                            0.00|                            0.00|                            1.00|
+```
+
+# AX650
+
+```
+/opt/test/clip # ./main -e compiled.axmodel -d onnx_models/decode.onnx -v vocab.
+txt -i images/ -t text.txt
+Engine creating handle is done.
+Engine creating context is done.
+Engine get io info is done.
+Engine alloc io is done.
+[I][                            init][ 275]: RGB MODEL
+
+per image:
+                 image path\text|                            bird|                             cat|                             dog|
+                 images/bird.jpg|                            1.00|                            0.00|                            0.00|
+                  images/cat.jpg|                            0.01|                            0.98|                            0.01|
+            images/dog-chai.jpeg|                            0.00|                            0.00|                            1.00|
+
+
+per text:
+                 text\image path|                 images/bird.jpg|                  images/cat.jpg|            images/dog-chai.jpeg|
+                            bird|                            1.00|                            0.00|                            0.00|
+                             cat|                            0.00|                            0.99|                            0.01|
+                             dog|                            0.00|                            0.00|                            1.00|
+
+```

src/Runner/CLIPAX650.hpp

@@ -24,6 +24,7 @@ class CLIPAX650 : public CLIP
             return;
         }
         cv::resize(image, input, cv::Size(224, 224));
+        cv::cvtColor(input, input, cv::COLOR_BGR2RGB);
         auto ret = m_encoder->inference();
 
         image_features.resize(LEN_IMAGE_FEATURE);

src/Runner/CLIPOnnx.hpp

@@ -26,6 +26,7 @@ class CLIPOnnx : public CLIP
             return;
         }
         cv::resize(image, input, cv::Size(224, 224));
+        cv::cvtColor(input, input, cv::COLOR_BGR2RGB);
 
         float *inputPtr = (float *)m_encoder->getInputPtr(0);
 

src/main.cpp

@@ -111,23 +111,27 @@ int main(int argc, char *argv[])
     mClip->decode(image_features, text_features, logits_per_image, logits_per_text);
 
     printf("\n");
-    printf("per image:\n");
-    printf("%32s|", "image path\\text");
-    for (size_t i = 0; i < texts.size(); i++)
+    if (texts.size() > 1)
     {
-        printf("%32s|", texts[i].c_str());
-    }
-    printf("\n");
-    for (size_t i = 0; i < logits_per_image.size(); i++)
-    {
-        printf("%32s|", image_paths[i].c_str());
-        for (size_t j = 0; j < logits_per_image[i].size(); j++)
+        printf("per image:\n");
+        printf("%32s|", "image path\\text");
+        for (size_t i = 0; i < texts.size(); i++)
         {
-            printf("%32.2f|", logits_per_image[i][j]);
+            printf("%32s|", texts[i].c_str());
+        }
+        printf("\n");
+        for (size_t i = 0; i < logits_per_image.size(); i++)
+        {
+            printf("%32s|", image_paths[i].c_str());
+            for (size_t j = 0; j < logits_per_image[i].size(); j++)
+            {
+                printf("%32.2f|", logits_per_image[i][j]);
+            }
+            printf("\n");
         }
         printf("\n");
     }
-    printf("\n");
+
     printf("\n");
     printf("per text:\n");
     printf("%32s|", "text\\image path");

text.txt

@@ -1,3 +1,3 @@
+bird
 cat
-dog
-brid
+dog