Updates:

- Update notebooks for recent Colab and Python versions.
- Update OD and Semseg documentation.
- Deprecate `allocate_inputs` and add new flags.
- Log errors for external tools.


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

Author: KBOUSTM

common/stm32ai_dc/types.py

@@ -1098,10 +1098,31 @@ class CliParameters(typing.NamedTuple):
     """ Optional: Compression level. Defaults to "lossless" """
 
     allocateInputs: bool = True
-    """ When set to true, activations buffer will be also used to handle the input buffers. Defaults to True """
+    """ DEPRECATED: Optional: When set to true, activations buffer will be also used to handle the input buffers. Defaults to True """
 
     allocateOutputs: bool = True
-    """ Optional: When set to true, activations buffer will be also used to handle the output buffers. Defaults to True """
+    """ DEPRECATED: Optional: When set to true, activations buffer will be also used to handle the output buffers. Defaults to True """
+
+    noInputsAllocation: Optional[bool] = None
+    """ input buffers must be provided by the application.
+        (default: input buffers are stored in a reserved space within the "activations" buffer) 
+    """
+    
+    noOutputsAllocation: Optional[bool] = None
+    """  output buffers must be provided by the application.
+        (default: output buffers are stored in a reserved space within the "activations" buffer) 
+    """
+    
+    inputDataType: Optional[str] = None
+    """ indicate the expected inputs data type of the generated model
+        Multiple inputs: in_data_type_1,in_data_type_2,...
+        If one data type is given, it will be applied for all inputs
+    """
+    outputDataType: Optional[str] = None
+    """indicate the expected outputs data type of the generated model
+        Multiple outputs: out_data_type_1,out_data_type_2,...
+        If one data type is given, it will be applied for all outputs
+    """
 
     target: Optional[str] = None
     """ Optional: Defines a serie which will be used to calculate more accurate memory footprint. Defaults to "stm32f4" """
@@ -1488,4 +1509,4 @@ def delete_validation_output_file(self, fileName: str):
 
     def delete_generated_file(self, fileName: str):
         raise NotImplementedError("delete_generated_file not implemented on \
-            this backend")
+            this backend")

common/stm32ai_local/compile.py

@@ -208,10 +208,12 @@ def cmd_compile(
     logger.info(' target      : %s', str(session.board))
     logger.info(' options     : %s', str(cmd_line[-1]))
 
-    err, _ = run_shell_cmd(cmd_line, logger=logger)
+    err, errorList = run_shell_cmd(cmd_line, logger=logger)
 
     if err != 0:
-        raise Exception('WRONG!')
+        for i in range(len(errorList)):
+            logger.info('%s', str(errorList[i]))
+        raise Exception('Error during compilation')
 
     # post-process the results
     if err == 0:

semantic_segmentation/datasets/pascal_voc_2012/README.md

@@ -22,7 +22,7 @@ The augmented dataset increases the number of images and masks to a total of 105
 
 The additional masks can be downloaded from the following Dropbox link:
 
-[Download Additional Masks](https://www.dropbox.com/path/to/masks)
+[Download Additional Masks](https://www.dropbox.com/scl/fi/xccys1fus0utdioi7nj4d/SegmentationClassAug.zip?rlkey=0wl8iz6sc40b3qf6nidun4rez&e=3&dl=0)
 
 These masks are integer masks with indices representing different classes. They should be placed in the same directory structure as the original Pascal VOC 2012 dataset.
 

semantic_segmentation/src/README.md

@@ -150,13 +150,8 @@ dataset:
 ```
 The `name` attribute is mandatory. To date, the unique accepted values without the `class_names`: `pascal_voc`, `indoor_pascal_voc`, `person_pascal_voc`, `outdoor_pascal_voc`. However, if using a different dataset and naming convention, the `class_name` attribute should be provided.
 
-**Please follow the [PASCAL VOC 2012 tutorial](../datasets/pascal_voc_2012/README.md) to have more training masks (about 10,582) and a `trainaug.txt` file containing the IDs of the new training masks. Or, the [COCO 2017 PASCAL VOC 2012 tutorial](../datasets/coco_2017_pascal_voc_2012/README.md) to create COCO 2017 + PASCAL VOC 2012 dataset, and the [n_class_coco_2017_pascal_voc_2012 tutorial](../datasets/n_class_coco_2017_pascal_voc_2012/README.md) to create the indoor_pascal_voc, outdoor_pascal_voc, person_pascal_voc dataset.**
-
----
-
 The `name` attribute is mandatory. To date, a unique value is accepted: `pascal_voc`. This parameter is used in the data_loader in order to correctly construct each dataset expecting [PASCAL VOC 2012](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/) structure. In the future, it could be extended to other dataset types.
 
-**Please follow the [PASCAL VOC 2012 tutorial](../datasets/pascal_voc_2012/README.md) to have more training masks (about 10,582) and a `trainaug.txt` file containing the IDs of the new training masks. Or, the [COCO 2017 PASCAL VOC 2012 tutorial](../datasets/coco_2017_pascal_voc_2012/README.md) to create COCO 2017 + PASCAL VOC 2012 dataset.**
 
 To adhere to the PASCAL VOC 2012 structure, we need to establish three paths for each set. This setup will automatically retrieve the specified images from the training path's JPEG images directory and the corresponding masks from the training masks path.
 
@@ -168,6 +163,8 @@ the validation set. The `seed` attribute specifies the seed value to use for ran
 The 3 quantization path attributes are used to specify a dataset for the quantization process. If these attributes are not provided and a training set is available, the training set is used for the quantization. 
 However, training sets can be quite large and the quantization process can take a long time to run. To avoid this issue, you can set the `quantization_split` attribute to use only a random portion of the dataset for quantization.
 
+**Please follow the [PASCAL VOC 2012 tutorial](../datasets/pascal_voc_2012/README.md) to have more training masks (about 10,582) and a `trainaug.txt` file containing the IDs of the new training masks. Or, the [COCO 2017 PASCAL VOC 2012 tutorial](../datasets/coco_2017_pascal_voc_2012/README.md) to create COCO 2017 + PASCAL VOC 2012 dataset, and the [n_class_coco_2017_pascal_voc_2012 tutorial](../datasets/n_class_coco_2017_pascal_voc_2012/README.md) to create the indoor_pascal_voc, outdoor_pascal_voc, person_pascal_voc dataset.**
+
 </details></ul>
 <ul><details open><summary><a href="#2-4">2.4 Apply image preprocessing</a></summary><a id="2-4"></a>