Release AI-ModelZoo-2.1.1:

- Fixed some bugs.
- Added more detailed information in README files.
- Updated development cloud benchmark notebook to include MPU boards.


Signed-off-by: khaoula boutiche <[email protected]>

Author: KBOUSTM

README.md

@@ -31,7 +31,7 @@ to train and deploy the models.
 </details>
 <details open><summary><b>2.1:</b></summary>
 
-* Included additional models compatible with the `STM32MP257F-DK2` board.
+* Included additional models compatible with the [STM32MP257F-EV1](https://www.st.com/en/evaluation-tools/stm32mp257f-ev1) board.
 * Added support for per-tensor quantization.
 * Integrated support for `ONNX model` quantization and evaluation.
 * Included support for `STEdgeAI` (STM32Cube.AI v9.1.0 and subsequent versions).
@@ -59,7 +59,7 @@ to train and deploy the models.
 | [MobileNet v1 0.25](image_classification/pretrained_models/mobilenetv1/README.md)   | 96x96x1<br> 96x96x3<br> 224x224x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   |
 | [MobileNet v1 0.5](image_classification/pretrained_models/mobilenetv1/README.md)   | 224x224x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   | 
 | [MobileNet v2 0.35](image_classification/pretrained_models/mobilenetv2/README.md)   | 128x128x3<br>  224x224x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   |
-| [MobileNet v2 1.0](image_classification/pretrained_models/mobilenetv2/README.md)   |  224x224x3     | Full IC Services      |  STM32MP257F-DK2<br>   |
+| [MobileNet v2 1.0](image_classification/pretrained_models/mobilenetv2/README.md)   |  224x224x3     | Full IC Services      |  [STM32MP257F-EV1](./X-LINUX-AI_application_code/image_classification/STM32MP2/README.md)<br>   |
 | [ResNet8 v1](image_classification/pretrained_models/resnetv1/README.md)   | 32x32x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   |
 | [ST ResNet8](image_classification/pretrained_models/resnetv1/README.md)   | 32x32x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   |
 | [ResNet32 v1](image_classification/pretrained_models/resnetv1/README.md)   | 32x32x3     | Full IC Services      | [STM32H747I-DISCO](stm32ai_application_code/image_classification/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> [NUCLEO-H743ZI2](stm32ai_application_code/image_classification/Application/NUCLEO-H743ZI2) with B-CAMS-OMV camera daughter board<br>   |
@@ -79,8 +79,8 @@ to train and deploy the models.
 | Models             | Input Resolutions | Supported Services    | Targets for deployment |
 |--------------------|------------------|-----------------------|-------------------|
 | [ST SSD MobileNet v1 0.25](object_detection/pretrained_models/st_ssd_mobilenet_v1/README.md)   |  192x192x3<br> 224x224x3<br> 256x256x3<br>  | Full OD Services      | [STM32H747I-DISCO](stm32ai_application_code/object_detection/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br>    |
-| [SSD MobileNet v2 fpn lite 0.35](object_detection/pretrained_models/ssd_mobilenet_v2_fpnlite/README.md)   |  192x192x3<br> 224x224x3<br> 256x256x3<br> 416x416x3   | Full OD Services      | [STM32H747I-DISCO](stm32ai_application_code/object_detection/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> STM32MP257F-DK2 <br>    |
-| [SSD MobileNet v2 fpn lite 1.0](object_detection/pretrained_models/ssd_mobilenet_v2_fpnlite/README.md)   |  256x256x3<br> 416x416x3   | Full OD Services      |  STM32MP257F-DK2   |
+| [SSD MobileNet v2 fpn lite 0.35](object_detection/pretrained_models/ssd_mobilenet_v2_fpnlite/README.md)   |  192x192x3<br> 224x224x3<br> 256x256x3<br> 416x416x3   | Full OD Services      | [STM32H747I-DISCO](stm32ai_application_code/object_detection/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br> or <br>[STM32MP257F-EV1](./X-LINUX-AI_application_code/object_detection/STM32MP2/README.md) <br>    |
+| [SSD MobileNet v2 fpn lite 1.0](object_detection/pretrained_models/ssd_mobilenet_v2_fpnlite/README.md)   |  256x256x3<br> 416x416x3   | Full OD Services      |  [STM32MP257F-EV1](./X-LINUX-AI_application_code/object_detection/STM32MP2/README.md)   |
 | [ST Yolo LC v1](object_detection/pretrained_models/st_yolo_lc_v1/README.md)   |  192x192x3<br> 224x224x3<br> 256x256x3<br>  | Full OD Services      | [STM32H747I-DISCO](stm32ai_application_code/object_detection/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br>    |
 | [Tiny Yolo v2](object_detection/pretrained_models/tiny_yolo_v2/README.md)   |  224x224x3<br> 416x416x3<br>  | Full OD Services      | [STM32H747I-DISCO](stm32ai_application_code/object_detection/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br>    |
 
@@ -93,20 +93,20 @@ to train and deploy the models.
 [Pose Estimation use case](pose_estimation)
 | Models             | Input Resolutions | Supported Services    | Targets for deployment |
 |--------------------|------------------|-----------------------|-------------------|
-| [Yolo v8 pose](pose_estimation/pretrained_models/yolov8_pose/README.md)   |  256x256x3<br>  | Evaluation / Benchmarking / Prediction / Deployment      | STM32MP257F-DK2 <br>  |
-| [MoveNet 17 kps](pose_estimation/pretrained_models/movenet/README.md)   |  192x192x3<br> 256x256x3<br>   | Evaluation / Quantization / Benchmarking / Prediction      | [STM32H747I-DISCO](stm32ai_application_code/pose_estimation/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br>|
-| [ST MoveNet 13 kps](pose_estimation/pretrained_models/movenet/README.md)   |  192x192x3<br>   | Evaluation / Quantization / Benchmarking / Prediction      | [STM32H747I-DISCO](stm32ai_application_code/pose_estimation/Application/STM32H747I-DISCO) with B-CAMS-OMV camera daughter board<br>|
+| [Yolo v8 n pose](pose_estimation/pretrained_models/yolov8n_pose/README.md)   |  256x256x3<br>  | Evaluation / Benchmarking / Prediction / Deployment      | [STM32MP257F-EV1](./X-LINUX-AI_application_code/pose_estimation/STM32MP2/README.md) <br>  |
+| [MoveNet 17 kps](pose_estimation/pretrained_models/movenet/README.md)   |  192x192x3<br> 224x224x3<br> 256x256x3<br>   | Evaluation / Quantization / Benchmarking / Prediction      | N/A <br>|
+| [ST MoveNet 13 kps](pose_estimation/pretrained_models/movenet/README.md)   |  192x192x3<br>   | Evaluation / Quantization / Benchmarking / Prediction      | N/A <br>|
 
 </details>
 
 <details open><summary><b>Segmentation (Seg)</b></summary>
 
-[Segmentation use case](segmentation)
+[Segmentation use case](./semantic_segmentation/)
 | Models             | Input Resolutions | Supported Services    | Targets for deployment |
 |--------------------|------------------|-----------------------|-------------------|
-| [DeepLab v3](segmentation/pretrained_models/deeplab_v3/README.md)   |  512x512x3<br>  | Full Seg Services     | STM32MP257F-DK2 <br>  |
+| [DeepLab v3](./semantic_segmentation/pretrained_models/deeplab_v3/README.md)   |  512x512x3<br>  | Full Seg Services     | [STM32MP257F-EV1](./X-LINUX-AI_application_code/semantic_segmentation/STM32MP2/README.md) <br>  |
 
-[Full Seg Services](segmentation/README.md) : training, evaluation, quantization, benchmarking, prediction, deployment
+[Full Seg Services](./semantic_segmentation/README.md) : training, evaluation, quantization, benchmarking, prediction, deployment
 
 </details>
 

X-LINUX-AI_application_code/pose_estimation/STM32MP2/README.md

@@ -45,7 +45,7 @@ For STM32MPU, application code example is provided as python script to facilitat
 
 You should use the deploy.py script to automatically deploy the program on the target (if the hardware is connected to the network).
 
-The [deployment script](../../../pose_estimation/deployment/README.md) of the model zoo is used to directly populate the target with all the needed files, resources and NN model. The script use
+The [deployment script](../../../pose_estimation/deployment/README_MPU.md) of the model zoo is used to directly populate the target with all the needed files, resources and NN model. The script use
 the IP address provided in the configuration yaml file. The application source code is available `Application\` directory and can be modified easily.
 
 ### __Deploy NN model__

audio_event_detection/src/user_config.yaml

@@ -10,7 +10,7 @@ general:
 
 operation_mode: chain_tbqeb #chain_tbqeb
 #choices=['training' , 'evaluation', 'prediction', 'deployment', 'quantization', 'benchmarking',
-#        'chain_tbqeb','chain_tqe',' chain_eqe','chain_qb','chain_eqeb','chain_qd ']
+#        'chain_tbqeb','chain_tqe','chain_eqe','chain_qb','chain_eqeb','chain_qd ']
 
 dataset:
   name: esc10

common/benchmarking/common_benchmark.py

@@ -178,9 +178,9 @@ def analyze_footprints(offline: bool = True, results: dict = None, stm32ai_outpu
             log_file.write(f'Model Name : {model_name} \n' + f'Tool version : {version_strings} \n'+ f'Model Size : {model_size} KiB \n' + f'Peak Ram usage : {peak_ram} KiB\n')
         if inference_res:
             inference_time = round(results["duration_ms"], 2)
-            npu_percent = 100*results["npu_percent"]
-            gpu_percent = 100*results["gpu_percent"]
-            cpu_percent = 100*results["cpu_percent"]
+            npu_percent = round(100*results["npu_percent"],2)
+            gpu_percent = round(100*results["gpu_percent"],2)
+            cpu_percent = round(100*results["cpu_percent"],2)
             print("[INFO] : Inference Time : {} (ms)".format(inference_time))
             print("[INFO] : Execution engine repartition : ")
             print("[INFO] :     NPU usage : {}".format(npu_percent))

common/evaluation/onnx_evaluation.py

@@ -5,16 +5,10 @@
 #  * the root directory of this software component.
 #  * If no LICENSE file comes with this software, it is provided AS-IS.
 #  *--------------------------------------------------------------------------------------------*/
-import onnxruntime
 import numpy as np
 import os
-from typing import Tuple, Optional, List, Dict
 import onnx
-import sys
 
-from visualize_utils import plot_confusion_matrix
-from sklearn.metrics import accuracy_score, confusion_matrix
-import mlflow
 
 def predict_onnx(sess: onnx.ModelProto, data: np.ndarray) -> np.ndarray:
     """
@@ -29,7 +23,15 @@ def predict_onnx(sess: onnx.ModelProto, data: np.ndarray) -> np.ndarray:
     """
     input_name = sess.get_inputs()[0].name
     label_name = sess.get_outputs()[0].name
-    onx_pred = sess.run([label_name], {input_name: data.astype(np.float32)})[0]
+    # Process each image individually
+    predictions = []
+    for i in range(data.shape[0]):
+        single_data = data[i:i + 1]  # Set batch size to 1
+        single_predictions = sess.run([label_name], {input_name: single_data.astype(np.float32)})[0]
+        predictions.append(single_predictions)
+    
+    # Concatenate all predictions
+    onx_pred = np.concatenate(predictions, axis=0)
     return onx_pred
 
 

common/onnx_utils/st_ssd_mobilenet_v1_025_256.onnx renamed to common/onnx_utils/sample_models/st_ssd_mobilenet_v1_025_256.onnx

@@ -20,9 +20,10 @@
 
 warnings.filterwarnings("ignore")
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
+import onnxruntime
 import tensorflow as tf
 import tqdm
-from typing import Optional
+from sklearn.metrics import accuracy_score, confusion_matrix
 
 from preprocess import apply_rescaling, postprocess_output, preprocess_input
 from visualize_utils import plot_confusion_matrix
@@ -157,8 +158,6 @@ def evaluate_onnx_model(input_samples:np.ndarray,
     prd_labels = predict_onnx(sess, input_samples).argmax(axis=1)
     val_acc = round(accuracy_score(input_labels, prd_labels) * 100, 2)
     print(f'[INFO] : Evaluation accuracy: {val_acc} %')
-    if not os.path.exists("outputs"):
-        os.makedirs("outputs")
     log_file_name = f"{output_dir}/stm32ai_main.log"
     with open(log_file_name, 'a', encoding='utf-8') as f:
         f.write(f'{model_type} onnx model\nEvaluation accuracy: {val_acc} %\n')

image_classification/src/evaluation/evaluate.py

@@ -9,7 +9,7 @@ general:
 
 operation_mode: chain_tbqeb
 #choices=['training' , 'evaluation', 'prediction', 'deployment', 'quantization', 'benchmarking',
-#        'chain_tbqeb','chain_tqe',' chain_eqe','chain_qb','chain_eqeb','chain_qd ']
+#        'chain_tbqeb','chain_tqe','chain_eqe','chain_qb','chain_eqeb','chain_qd ']
 
 dataset:
   name: flowers

image_classification/src/user_config.yaml

@@ -210,7 +210,17 @@ def parse_data_augmentation_section(cfg: DictConfig, config_dict: Dict) -> None:
             cfg.data_augmentation.config = config_dict['custom_data_augmentation']['config'].copy()
         del cfg.custom_data_augmentation
 
-def get_class_names_from_file(cfg: DictConfig) -> List:
+def get_class_names_from_file(cfg: DictConfig) -> List[str]:
+    """
+    Reads class names from a file specified in the configuration and returns them as a list.
+    Args:
+        cfg (DictConfig): Configuration object containing the path to the label file.
+    Returns:
+        List[str]: A list of class names read from the file.
+    Raises:
+        FileNotFoundError: If the file specified in the configuration does not exist.
+        IOError: If there is an error reading the file.
+    """
     if cfg.deployment.label_file_path :
         with open(cfg.deployment.label_file_path, 'r') as file:
             class_names = [line.strip() for line in file]
@@ -396,21 +406,13 @@ def get_config(config_data: DictConfig) -> DefaultMunch:
 
     # Check that all datasets have the required directory structure
     cds = cfg.dataset
-    if not cds.class_names and cfg.operation_mode not in ("quantization", "benchmarking", "chain_qb"):
+    if not cds.class_names and cfg.operation_mode not in ("benchmarking"):
         # Infer the class names from a dataset
         for path in [cds.training_path, cds.validation_path, cds.test_path, cds.quantization_path]:
             if path:
                 cds.class_names = get_class_names(dataset_root_dir=path)
                 print("[INFO] : Found {} classes in dataset {}".format(len(cds.class_names), path))
-                break
-
         if not cds.class_names and cfg.operation_mode in ("deployment","chain_qd") and cfg.hardware_type == "MPU":
             cds.class_names = get_class_names_from_file(cfg)
             print("[INFO] : Found {} classes in label file {}".format(len(cds.class_names), cfg.deployment.label_file_path))
-
-        # This should not happen. Just in case.
-        if not cds.class_names:
-            raise ValueError("\nMissing `class_names` attribute\nPlease check the 'dataset' section of your "
-                             "configuration file.")
-
     return cfg

image_classification/src/utils/parse_config.py

@@ -9,7 +9,7 @@ general:
 operation_mode: chain_tqeb
 dataset:
   name: coco_80_classes
-  class_names:['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
+  class_names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
   training_path: DATASETS/train_set
   validation_path: DATASETS/val_set
   test_path: