add large model

README.md

@@ -13,8 +13,6 @@ tags:
 
 This project only to show a demo of Bird Species Classification model within 1400+ species of birds.
 
-The model is trained with 224x224 resolution.
-
 This model has been converted to run on the Axera NPU using **w8a16** quantization.
 
 This model has been optimized with the following LoRA: 
@@ -23,6 +21,11 @@ Compatible with Pulsar2 version: 5.1
 
 ## Convert tools links:
 
+Do model convert from onnx to axmodel with commands like:
+```
+pulsar2 build --config ./quant/bird-l.json
+```
+
 For those who are interested in model conversion, you can try to export axmodel through 
 
 - [Pulsar2 Link, How to Convert ONNX to axmodel](https://pulsar2-docs.readthedocs.io/en/latest/pulsar2/introduction.html) 
@@ -38,39 +41,62 @@ For those who are interested in model conversion, you can try to export axmodel
   - [Module-LLM](https://docs.m5stack.com/zh_CN/module/Module-LLM)
   - [LLM630 Compute Kit](https://docs.m5stack.com/zh_CN/core/LLM630%20Compute%20Kit)
  
-| Models       | Platforms    | latency       | Top5 Accuracy | CMM size(MB)  |
-| -------------| -------------| ------------- | --------------| --------------|
-|              | AX650        | 0.19ms        |               |               |
-| bird-s       | AX630C       | 0.54ms        | 66%           | 1.07          |
-|              | AX615        | 0.87ms        |               |               |
-|              | AX650        | 0.58ms        |               |               |
-| bird-m       | AX630C       | 2.52ms        | 79%           | 12.2          |
-|              | AX615        | 4.97ms        |               |               |
+| Models       | Platforms    | latency      | Top1 Accuracy | Top5 Accuracy | CMM size(MB)  |
+| -------------| -------------| -------------| --------------| --------------| --------------|
+|              | AX650        | 0.19ms       |               |               |               |
+| bird-s       | AX630C       | 0.54ms       | 44%           | 66%           | 1.07          |
+|              | AX615        | 0.87ms       |               |               |               |
+|              | AX650        | 0.58ms       |               |               |               |
+| bird-m       | AX630C       | 2.52ms       | 59%           | 79%           | 12.2          |
+|              | AX615        | 4.97ms       |               |               |               |
+|              | AX650        | 5.60ms       |               |               |               |
+| bird-l       | AX630C       | 35.2ms       | 86%           | 95%           | 29.6          |
+|              | AX615        | 64.1ms       |               |               |               |
 
 ## How to use
 
 Download all files from this repository to the device
 
 ```
-root@ax650:~/Bird-Species-Classification# tree
+(base) root@ax650:~/Bird-Species-Classification# tree
 .
 ├── README.md
 ├── axmodel_infer.py
 ├── class_name.txt
+├── config.json
 ├── model
-│   ├── AX615
-│   │   ├── bird_615_npu1.axmodel
-│   │   └── bird_615_npu2.axmodel
-│   ├── AX620E
-│   │   ├── bird_630_npu1.axmodel
-│   │   └── bird_630_npu2.axmodel
-│   └── AX650
-│       └── bird_650_npu3.axmodel
+│   ├── bird-l
+│   │   ├── AX615
+│   │   │   └── bird_615_npu2.axmodel
+│   │   ├── AX620E
+│   │   │   ├── bird_630_npu1.axmodel
+│   │   │   └── bird_630_npu2.axmodel
+│   │   └── AX650
+│   │       └── bird_650_npu3.axmodel
+│   ├── bird-m
+│   │   ├── AX615
+│   │   │   ├── bird_615_npu1.axmodel
+│   │   │   └── bird_615_npu2.axmodel
+│   │   ├── AX620E
+│   │   │   ├── bird_630_npu1.axmodel
+│   │   │   └── bird_630_npu2.axmodel
+│   │   └── AX650
+│   │       └── bird_650_npu3.axmodel
+│   └── bird-s
+│       ├── AX615
+│       │   ├── bird_615_npu1.axmodel
+│       │   └── bird_615_npu2.axmodel
+│       ├── AX630C
+│       │   ├── bird_630_npu1.axmodel
+│       │   └── bird_630_npu2.axmodel
+│       └── AX650
+│           └── bird_650_npu3.axmodel
 ├── onnx_infer.py
 ├── prediction_result_top5.png
 ├── quant
-│   ├── Bird.json
-│   ├── README.md
+│   ├── bird-l.json
+│   ├── bird-m.json
+│   ├── bird-s.json
 │   └── bird.tar.gz
 └── test_images
     ├── 03111_2c0dfa5a-c4a0-47f8-ac89-6a289208050f.jpg
@@ -81,7 +107,7 @@ root@ax650:~/Bird-Species-Classification# tree
     ├── 04405_0c5a6785-0bc2-49d9-9702-b9e94ba9b686.jpg
     └── 04593_3d74d5a7-15b1-4bb9-af6f-1bcd78485787.jpg
 
-6 directories, 20 files
+15 directories, 31 files
 
 ```
 
@@ -96,27 +122,46 @@ wget https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3rc0/axengin
 pip install axengine-0.1.3-py3-none-any.whl
 ```
 
+## Inference with ONNX model
+```
+root@ebba5440b03c:/home/# python onnx_infer.py -m Birdmodel_inat_bird-l.onnx --image test_images/04251_3a52191e-be71-4539-98ea-14a8f2347330.jpg
+build predictor with Birdmodel_inat_bird-l.onnx...
+Loading ONNX model with providers: ['CPUExecutionProvider']
+
+Image: test_images/04251_3a52191e-be71-4539-98ea-14a8f2347330.jpg
+Top-5 Predictions:
+#1: 04251_Animalia_Chordata_Aves_Passeriformes_Tityridae_Tityra_semifasciata (0.9231)
+#2: 04019_Animalia_Chordata_Aves_Passeriformes_Oriolidae_Sphecotheres_vieilloti (0.0021)
+#3: 03233_Animalia_Chordata_Aves_Anseriformes_Anatidae_Cairina_moschata (0.0006)
+#4: 04219_Animalia_Chordata_Aves_Passeriformes_Thraupidae_Paroaria_capitata (0.0004)
+#5: 03912_Animalia_Chordata_Aves_Passeriformes_Laniidae_Lanius_minor (0.0003)
+Result saved to: prediction_result_top5.png
+
+```
+
 ## Inference with AX650 Host, such as M4N-Dock(爱芯派Pro)
 
 ```
-root@ax650:~/Bird-Species-Classification# python3 axmodel_infer.py --image test_images/04251_3a52191e-be71-4539-98ea-14a8f2347330.jpg
+root@ax650:~/bird# python3 axmodel_infer.py -m bird_650_npu3.axmodel -i test_images/04251_3a52191e-be71-4539-98ea-14a8f2347330.jpg
 [INFO] Available providers:  ['AxEngineExecutionProvider']
+build predictor with bird_650_npu3.axmodel...
 Loading ONNX model with providers: ['AxEngineExecutionProvider']
 [INFO] Using provider: AxEngineExecutionProvider
 [INFO] Chip type: ChipType.MC50
 [INFO] VNPU type: VNPUType.DISABLED
 [INFO] Engine version: 2.12.0s
 [INFO] Model type: 2 (triple core)
-[INFO] Compiler version: 5.1-patch1 74996179
+[INFO] Compiler version: 5.1-patch1 8c5871d5
 
 Image: test_images/04251_3a52191e-be71-4539-98ea-14a8f2347330.jpg
 Top-5 Predictions:
-#1: 04251_Animalia_Chordata_Aves_Passeriformes_Tityridae_Tityra_semifasciata (0.9823)
-#2: 04333_Animalia_Chordata_Aves_Passeriformes_Tyrannidae_Tyrannus_savana (0.0103)
-#3: 04208_Animalia_Chordata_Aves_Passeriformes_Sylviidae_Sylvia_melanocephala (0.0056)
-#4: 03561_Animalia_Chordata_Aves_Coraciiformes_Alcedinidae_Todiramphus_chloris (0.0009)
-#5: 04455_Animalia_Chordata_Aves_Piciformes_Picidae_Melanerpes_lewis (0.0005)
+#1: 04251_Animalia_Chordata_Aves_Passeriformes_Tityridae_Tityra_semifasciata (0.9137)
+#2: 04019_Animalia_Chordata_Aves_Passeriformes_Oriolidae_Sphecotheres_vieilloti (0.0022)
+#3: 03233_Animalia_Chordata_Aves_Anseriformes_Anatidae_Cairina_moschata (0.0006)
+#4: 03912_Animalia_Chordata_Aves_Passeriformes_Laniidae_Lanius_minor (0.0005)
+#5: 04219_Animalia_Chordata_Aves_Passeriformes_Thraupidae_Paroaria_capitata (0.0004)
 Result saved to: prediction_result_top5.png
+
 ```
 
 output：

axmodel_infer.py

@@ -21,16 +21,17 @@ plt.rcParams['axes.unicode_minus'] = False
 
 class BirdPredictorONNX:
     """Bird classification predictor based on ONNX Runtime"""
-    
-    def __init__(self, class_name_file, model_file):
+
+    def __init__(self, class_name_file, model_file, mean, std, image_size=224):
         """
         Initialize the predictor.
         Defaults to AxEngineExecutionProvider.
         """
-        self.rgb_mean = [0.5,0.5,0.5]
-        self.rgb_std = [0.5,0.5,0.5]
+        self.rgb_mean = mean
+        self.rgb_std = std
+        self.image_size = image_size
         self.classes = self.load_classes(class_name_file)
-        
+        print(f"build predictor with {model_file}...")
         providers = ['AxEngineExecutionProvider']
         print(f"Loading ONNX model with providers: {providers}")
         
@@ -57,8 +58,8 @@ class BirdPredictorONNX:
     
     def preprocess_image(self, image_path):
         image = Image.open(image_path).convert('RGB')
-        image = image.resize((224, 224), Image.BILINEAR)
-        
+        image = image.resize((int(self.image_size), int(self.image_size)), Image.BICUBIC)
+
         img_array = np.array(image, dtype=np.uint8)
         img_array = img_array.transpose(2, 0, 1)
         img_array = np.expand_dims(img_array, axis=0)
@@ -226,12 +227,25 @@ def main():
                        default="./class_name.txt",
                        help="Path to configuration file")
     parser.add_argument("-m", "--model_file", 
-                       default="./model/AX650/bird_650_npu3.axmodel", 
+                       default="./bird_650_npu3.axmodel", 
                        help="Path to ONNX model file")
+    parser.add_argument("-imgsz", "--image_size", 
+                       default=384, 
+                       help="Input image size")
+    parser.add_argument("-mean", "--mean", 
+                       type=float, 
+                       nargs='+', 
+                       default=[0.485, 0.456, 0.406], 
+                       help="Mean normalization values")
+    parser.add_argument("-std", "--std", 
+                       type=float, 
+                       nargs='+', 
+                       default=[0.229, 0.224, 0.225], 
+                       help="Standard deviation normalization values")
     parser.add_argument("--image_dir", 
                        default="./test_images",
                        help="Directory containing test images")
-    parser.add_argument("--image", 
+    parser.add_argument("-i", "--image", 
                        help="Path to a single test image")
     parser.add_argument("--top_k",
                        type=int,
@@ -239,10 +253,10 @@ def main():
                        help="Number of top predictions to show (default: 5)")
     
     args = parser.parse_args()
-    
-    
-    predictor = BirdPredictorONNX(args.class_map_file, args.model_file)
-    
+
+
+    predictor = BirdPredictorONNX(args.class_map_file, args.model_file, args.mean, args.std, args.image_size)
+
     if args.image and os.path.exists(args.image):
         try:
             top_k_results = predictor.predict_image_topk(args.image, k=args.top_k)

model/bird-l/AX615/bird_615_npu2.axmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8d71be9d5145d0256465968e0d15cf2de585d6c44fd545be33bd3c663feda29
+size 23482611

model/bird-l/AX620E/bird_630_npu1.axmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12d4e9dd34ffa17c4c06ea35883855234adcee8b243d8cc40d3d11025de508db
+size 24419641

model/bird-l/AX620E/bird_630_npu2.axmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70cb2a12d16c05d6e186af8b656c0209bfb5c62de67fd7efc5006620683712e6
+size 24269265

model/bird-l/AX650/bird_650_npu3.axmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfdb20082d3972977699a6fdf1c97100016cdcccbd9591462cc2095e63ad1c04
+size 23236891

onnx_infer.py

@@ -21,16 +21,17 @@ plt.rcParams['axes.unicode_minus'] = False
 
 class BirdPredictorONNX:
     """Bird classification predictor based on ONNX Runtime"""
-    
-    def __init__(self, class_name_file, model_file):
+
+    def __init__(self, class_name_file, model_file, mean, std, image_size=224):
         """
         Initialize the predictor.
         Defaults to CPUExecutionProvider.
         """
-        self.rgb_mean = [0.5,0.5,0.5]
-        self.rgb_std = [0.5,0.5,0.5]
+        self.rgb_mean = mean
+        self.rgb_std = std
+        self.image_size = image_size
         self.classes = self.load_classes(class_name_file)
-        
+        print(f"build predictor with {model_file}...")
         providers = ['CPUExecutionProvider']
         print(f"Loading ONNX model with providers: {providers}")
         
@@ -57,7 +58,7 @@ class BirdPredictorONNX:
     
     def preprocess_image(self, image_path):
         image = Image.open(image_path).convert('RGB')
-        image = image.resize((224, 224), Image.BILINEAR)
+        image = image.resize((int(self.image_size), int(self.image_size)), Image.BICUBIC)
         
         img_array = np.array(image, dtype=np.float32) / 255.0
         img_array = img_array.transpose(2, 0, 1)
@@ -230,12 +231,25 @@ def main():
                        default="./class_name.txt",
                        help="Path to configuration file")
     parser.add_argument("-m", "--model_file", 
-                       default="./bird_resnet18.onnx", 
+                       default="./bird-l.onnx", 
                        help="Path to ONNX model file")
+    parser.add_argument("-imgsz", "--image_size", 
+                       default=384, 
+                       help="Input image size")
+    parser.add_argument("-mean", "--mean", 
+                       type=float, 
+                       nargs='+', 
+                       default=[0.485, 0.456, 0.406], 
+                       help="Mean normalization values")
+    parser.add_argument("-std", "--std", 
+                       type=float, 
+                       nargs='+', 
+                       default=[0.229, 0.224, 0.225], 
+                       help="Standard deviation normalization values")
     parser.add_argument("--image_dir", 
                        default="./test_images",
                        help="Directory containing test images")
-    parser.add_argument("--image", 
+    parser.add_argument("-i", "--image", 
                        help="Path to a single test image")
     parser.add_argument("--top_k",
                        type=int,
@@ -243,10 +257,10 @@ def main():
                        help="Number of top predictions to show (default: 5)")
     
     args = parser.parse_args()
-    
-    
-    predictor = BirdPredictorONNX(args.class_map_file, args.model_file)
-    
+
+
+    predictor = BirdPredictorONNX(args.class_map_file, args.model_file, args.mean, args.std, args.image_size)
+
     if args.image and os.path.exists(args.image):
         try:
             top_k_results = predictor.predict_image_topk(args.image, k=args.top_k)

quant/{Bird.json → bird-l.json}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:218c30ea230c8b830402bccced57c2d6727f0f2d9d14e61d4315d9ce2e2e79ae
-size 893
+oid sha256:d7a41de236d68ddfe7f15614ba8137fc7f7acbb891387d121b225d09f41a1240
+size 899

quant/bird-m.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f70945c77a5e91a6bc21eed005fbc8004464703c5472cac1753e99f234b53517
+size 886

quant/bird-s.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d9f55be272bb41c5d32136e3cff83778b48603076b2b6ce1c29a5a1f523e9ff
+size 899